Research Report

Question: What are the most significant developments in AI this week?

Date: 2026-08-26T13:28:42.604718090+00:00

Coverage window: 2026-08-20 – 2026-08-26

Rounds: 4

Status: COMPLETE

Evidence: 65 claims · 55 sourced · 5 partial · 1 unsupported · 3 self-reported (no independent source) · 1 single-source

Executive Summary

As of 2026-08-26, the most significant AI developments of the week (Aug 20–26) are three: the OX Alpha mystery resolved as a Z.AI (Zhipu) GLM model, Alibaba's Qwen3.8-Flash-Next open-weight release previewing the Qwen4 architecture, and OpenAI's first measured results for its Jalapeño custom inference chip. No frontier-lab flagship model launched in-window — GPT-5.6, Claude Opus 5/Sonnet 5, Gemini 3.7 Flash, and Grok 4.6 all shipped before Aug 20 — so the week's center of gravity was open weights, inference economics, business deals, and California policy.

The week, ranked:

1. OX Alpha → Z.AI. The anonymous free model that appeared on OpenRouter Aug 20 and blew past DeepSeek in usage was confirmed by Z.AI to Bloomberg on Aug 26 as a GLM-series iteration, with open weights promised that night (not yet published at last check). The viral "80% DeepSWE" figure was a 10-task informal sample; community full-benchmark runs cluster at 58–63% (OpenRouter, LLMs Blog).
2. Qwen3.8-Flash-Next (Aug 26). The first open-weight preview of the architecture that will underpin Qwen4: 125B main + 51B N-gram embeddings + 4B MTP (≈180B on disk) with only 6B active per token, GDN+QSA hybrid attention, 262K native context, trained at ~1/9 the cost of Qwen3.7-Plus (GitHub, Hugging Face).
3. OpenAI's Jalapeño chip (Aug 25). First measured results: 1.5–1.9× more AI work per watt, 1.7–3.6× lower end-to-end latency, and 2.1–4.1× higher performance on highly interactive workloads on the InferenceX benchmark; deployment into OpenAI's own infrastructure planned by end of 2026, Gen 2 in development (OpenAI).
4. Nvidia's week. The $7B Poolside deal (Aug 20) and Q2 FY2027 earnings after the close Aug 26 — the event is confirmed, but the actual figures and market reaction were not yet confirmable from any indexed source when research ended; Nvidia's own guidance had been $91B ±2% against Q1's $81.6B (Nvidia IR).
5. AI policy and safety. California's 24 AI bills near final passage (SB 1159 passed Aug 20; AB 1651 signed Aug 22); OpenAI reversed its earlier opposition and urged California to strengthen SB 53 (Aug 22); Guidelight found Anthropic, Google, OpenAI, Meta, and xAI lack concrete, verifiable rogue-model containment plans (Aug 22).
6. Anthropic and Google. Claude Memory now works across chat and Cowork with editable topics and a sensitive-topics opt-in (Aug 25); a $5M wellbeing-evaluations grant program (Aug 25); Google Research's AgentHands XR (Aug 25) and DeepMind's EVE Online research partnership (Aug 21).

This Week's AI Developments

Analysis

The week's two model stories — Qwen's 6B-active MoE at $0.16/M input and Z.AI's free one-week mega-preview — both point the same direction: frontier-class capability is commoditizing fast, and the battlefield is shifting to price, distribution, and open weights. OpenAI's Jalapeño results and "full stack" essay are the counter-move: own the silicon and the serving layer. Nvidia sits at the center of the buildout either way — the Poolside deal is a new template for buying model capabilities without a full acquisition, and the ~$91B revenue quarter it reported (figures pending independent confirmation) is the number the whole market priced this week. The $220B YTD AI-debt figure shows how much of this is being levered rather than equity-funded.

The policy thread is the other through-line. California is effectively regulating US AI this month: 24 bills are heading to Newsom, and OpenAI's reversal on SB 53 — from opponent to advocate of stronger state rules as a national-standard base — is a genuine industry signal. Against that, the Guidelight containment findings and a transatlantic study of 239 AI-enabled radiology products (median 17.5-month second-authorization delay for CE-first vs 3.5 months for FDA-first; npj Digital Medicine) both say the same thing: deployment is running ahead of safety and regulatory machinery.

What didn't happen is also informative. No OpenAI, Google/DeepMind, Anthropic, Meta, xAI, or DeepSeek flagship launched in-window — DeepSeek's newsroom has had zero posts since Apr 24 (deepseek.com/news). Microsoft shipped no MAI model or Copilot super-app merge in-window; the merge and feature retirements were announced Aug 13/18, before this week, and the only in-window Copilot items were Slack/Teams integrations (Aug 21) and a Customize-tab GA (Aug 25) (GitHub Changelog). Meta's only dated AI item was an Aug 25 NABTU skilled-trades workforce partnership (Meta Newsroom). No dated in-window source was found for federal US, EU AI Act enforcement, or AI court rulings.

Risks & Open Questions

• Nvidia's actual Q2 FY27 numbers are unverified. The release landed after US market close on Aug 26, hours before the final research capture, and no indexed source carried the figures yet. Revenue, data-center revenue, Q3 guidance, and the share reaction should be re-checked against the official release or the EDGAR 8-K (CIK 0001045810) — treat any circulating numbers as unconfirmed.
• Z.AI's Ox Alpha open weights were not yet published at the last check (~13:26 UTC Aug 26): no zai-org/GLM-5.3 or zai-org/Ox-Alpha repo existed, and a Z.AI org member answered only "soon." The exact checkpoint (GLM-5 vs 5.3 vs "GLM Flash") and license remain unconfirmed, and the OpenRouter route's post-Aug-26 fate (paid vs removed) is unobserved.
• Benchmark integrity: the 80% DeepSWE headline for OX Alpha was a 10-task sample; community full runs sit at ~58–63% with one contested ~80% claim. All Qwen3.8-Flash-Next scores are Alibaba self-reported with no independent run yet.
• License detail with commercial implications: Qwen3.8-Flash-Next uses qwen-community-1.0, not Apache-2.0 — a contrast with Qwen3.8-27B that the release hype did not highlight.

Claims without independent support

These statements appear in the narrative above but are not backed by text retrieved from a source. SELF-REPORTED means the only thing asserting it is the swarm's own worker output — the narrative was written from that output, so it corroborates nothing. Treat all of these as unverified.

• [PARTIAL] Qwen3.8-Flash-Next (Aug 26). The first open-weight preview of the architecture that will underpin Qwen4: 125B main + 51B N-gram embeddings + 4B MTP (≈180B on disk) with only 6B active per token, GDN+QSA hybrid attention, 262K native context, trained at ~1/9 the cost of Qwen3.7-Plus (GitHub, Hugging Face). (unmatched: 180)
• [PARTIAL] AI policy and safety. California's 24 AI bills near final passage (SB 1159 passed Aug 20; AB 1651 signed Aug 22); OpenAI reversed its earlier opposition and urged California to strengthen SB 53 (Aug 22); Guidelight found Anthropic, Google, OpenAI, Meta, and xAI lack concrete, verifiable rogue-model containment plans (Aug 22). (unmatched: 1159, 1651)
• [SELF-REPORTED] Nvidia Q2 FY27 earnings released after close (quarter ended Jul 26); figures not yet indexed at final capture
• [PARTIAL] California: SB 1159 final passage (AI agents excluded from public-records "person" definition, 37–0); AB 1651 signed (bar-exam AI disclosure, operative Jan 2028); 24 AI bills near final passage (unmatched: 1159, 1651)
• [PARTIAL] arXiv: Recuris (recursive experiential-working memory; +17.8 pts for GPT-5.6 Sol, +15.6 for Claude Opus 5 on tau-bench); LAION-BVD (80M videos / 10M hours, largest open video corpus); SPO++ agentic-RL method (unmatched: 17.8, 15.6)
• [UNSUPPORTED] http://arxiv.org/abs/2608.24876v1 ; http://arxiv.org/abs/2608.24845v1 (unmatched: 2608.2488, 2608.2485)
• [PARTIAL] Against that, the Guidelight containment findings and a transatlantic study of 239 AI-enabled radiology products (median 17.5-month second-authorization delay for CE-first vs 3.5 months for FDA-first; npj Digital Medicine) both say the same thing: deployment is running ahead of safety and regulatory machinery. (unmatched: 239, 17.5, 41746, 3165)
• [SELF-REPORTED] Nvidia's actual Q2 FY27 numbers are unverified. The release landed after US market close on Aug 26, hours before the final research capture, and no indexed source carried the figures yet.
• [SELF-REPORTED] Z.AI's Ox Alpha open weights were not yet published at the last check (~13:26 UTC Aug 26): no zai-org/GLM-5.3 or zai-org/Ox-Alpha repo existed, and a Z.AI org member answered only "soon." The exact checkpoint (GLM-5 vs 5.3 vs "GLM Flash") and license remain unconfirmed, and the OpenRouter route's post-Aug-26 fate (paid vs removed) is unobserved.

Detailed Findings

Round 0 · Finding 1

Most Significant AI Developments — Week of 2026-08-20 to 2026-08-26

Note on scope: This week's headline developments center on OpenAI (custom silicon + platform moves), a new open-weight flagship from Alibaba/Qwen at the very end of the window, and a product update from Anthropic. On the model front, the major frontier releases (GPT-5.6 family, Claude Opus 5/Sonnet 5, Gemini 3.7 Flash, Grok 4.6) all landed before this week's window — I flag them as background only. In-window model releases were comparatively light.

1. OpenAI — Jalapeño custom inference chip, first measured results (Aug 25, 2026)

OpenAI published first benchmark results for "Jalapeño," its first custom inference chip. Per OpenAI's engineering post, dated August 25, 2026:

• On the public InferenceX benchmark (SemiAnalysis), Jalapeño delivered ~1.5–1.9× more AI work per watt at peak throughput and 1.7–3.6× lower end-to-end latency vs. comparison systems across GPT-OSS 120B, DeepSeek R1 670B, and Kimi K2.5 1T, with 2.1–4.1× higher performance on highly interactive workloads.
• Chip rated at 700 W (sustained ≤550 W on tested workloads); compared against GB200/GB300.
• OpenAI says it moved from design to tapeout in ~9 months using AI, and plans to begin deploying Jalapeño in its own infrastructure by end of 2026 (Gen 2 in development, Gen 3 taking shape).
• OpenAI stressed it will continue using NVIDIA and other partners' accelerators.

Source: https://openai.com/index/jalapeno-first-results/ (dated Aug 25, 2026)

Companion strategy post (Aug 25, 2026): CFO Sarah Friar authored "The full stack behind abundant intelligence," framing Jalapeño within OpenAI's compute strategy and noting GPT-5.6 Sol with max reasoning set "a new high" on the Artificial Analysis Coding Agent Index while using 54% fewer output tokens than another leading model. It also names current partners including Microsoft, NVIDIA, AWS, AMD, Broadcom, Cerebras, CoreWeave, Oracle, SB Energy, and SoftBank. Source: https://openai.com/index/the-full-stack-behind-abundant-intelligence/ (dated Aug 25, 2026)

2. OpenAI — GPT-5.6 family now available in Kiro, the AWS software-dev agent (Aug 24, 2026)

OpenAI announced the GPT-5.6 model family (Sol, Terra, Luna) is available in Kiro, an AWS-associated software development agent. Dated August 24, 2026. OpenAI and AWS jointly optimized the environment; testing found GPT-5.6 Terra completed tasks successfully in Kiro at roughly an 82% cost reduction on Terminal-Bench 2.1. Sources include quotes from AWS VP Swami Sivasubramanian (Agentic AI) and OpenAI VP Colleen Kapase. Source: https://openai.com/index/gpt-5-6-in-kiro/ (dated Aug 24, 2026)

3. OpenAI — New "Admin plugin" for ChatGPT Work and Codex (Aug 25, 2026)

…(truncated — the summary above captures the substance)

Round 0 · Finding 2

AI Policy, Regulatory, Safety & Legal Developments — August 20–26, 2026

Scope note: This report covers the week 2026-08-20 through 2026-08-26 for AI policy, regulatory, safety, and legal events. Items dated just outside that window are flagged explicitly as BACKGROUND and are never used for in-window claims. Sources are all pages I fetched; dates are the publication/update dates shown on those pages.

1. US State Legislatures & Signed Laws

California — final passage week, 24 AI bills near the finish line

2026-08-21 — The Transparency Coalition's weekly AI legislative update (published 2026-08-21) reports that California lawmakers are scheduled to adjourn sine die at end of business Monday, Aug. 31, with 24 AI-related bills near final passage heading to Gov. Gavin Newsom's desk (whom they note has 30 days to sign/veto). Seven states remain in session through the week: California, Michigan, Pennsylvania, Massachusetts, Ohio, New Jersey, and North Carolina. Michigan lawmakers are scheduled to return to Lansing on Tuesday, Aug. 25. (Source: https://www.transparencycoalition.ai/news/ai-legislative-update-august21-2026)

California SB 1159 — final passage (AI agents excluded from public-records "person" definition)

2026-08-20 — Per the same Transparency Coalition update (Aug 21), California SB 1159 received final passage in the Senate (37–0) on Aug. 20. The bill specifies that, for the California Public Records Act and other open-meeting acts, terms such as "person," "interested person," "participant," and "member of the public" do not include AI systems, autonomous agents, robots, or other non-human entities, whether physical or digital. Final passage places it among the bills headed to the governor. (Source: https://www.transparencycoalition.ai/news/ai-legislative-update-august21-2026)

California AB 1651 — Governor signs AI bar-exam disclosure law

2026-08-22 — Gov. Gavin Newsom signed AB 1651 and it was filed with the Secretary of State the same day. The law requires the State Bar of California to disclose when AI-generated content is used in the development or administration of bar examinations (questions, answer materials, study resources), even when a human later reviews or modifies the content. The provisions become operative on January 1, 2028. The report notes the law's significance: shifting the threshold toward traceability of AI content origin rather than merely whether a human reviewed it. (Source: https://www.promptinjection.net/p/ai-llm-news-roundup-august-09-august-22-2026 — itself dated Aug 23, 2026, referencing California Legislative Information)

…(truncated — the summary above captures the substance)

Round 0 · Finding 3

Significant AI Business Developments — Week of 2026-08-20 to 2026-08-26

Executive Summary

The window (Aug 20–26, 2026) was dominated by one mega-deal in AI infrastructure — Nvidia's ~$7B strategic licensing-plus-investment deal with the coding-AI startup Poolside (announced Aug 20) — and by Nvidia's fiscal Q2 2027 earnings report (released after market close Aug 26). The biggest lab funding rounds of the year (OpenAI's $110B, Anthropic's $65B Series H, Cognition's $40B talks) all fall before this window and are treated as background only. A Reuters-backed datapoint on surging corporate AI debt issuance ($220B YTD in 2026, as of Aug 21) rounded out the financing picture.

Because the window is short (7 days) and lean on newly announced lab mega-rounds, the strongest in-window primary sourced events are the Nvidia-Poolside deal and the Nvidia earnings. Several secondary aggregates (techstartups.com Aug 21 digest, theneuron.ai Aug 21 digest) describe the week's broader business news but were not fully retrievable in readable form.

Key Findings (in-window, dated)

1. Nvidia–Poolside: $6B software license + $1B equity investment (ANNOUNCED AUG 20, 2026)

• On August 20, 2026, Nvidia announced a transaction with the AI coding startup Poolside structured as a ~$7B strategic partnership and technology-licensing agreement rather than an outright acquisition. Terms: $6 billion for a non-exclusive license to Poolside's "Model Factory" model-development platform, plus a $1 billion equity investment at a $12 billion pre-money valuation; Nvidia also extended job offers to 109 Poolside engineers/researchers to work on its open-weight Nemotron model project. (Source: opensourceforu.com — explicitly states "On August 20, 2026, Nvidia announced..." while the article itself is dated Aug 25, 2026, https://www.opensourceforu.com/2026/08/nvidia-signs-deal-with-poolside/)
• Corroborating coverage dated within/near the window: Forbes (Aug 24, 2026) reports the $6B license, $1B investment, and 109 hires (https://www.forbes.com/sites/jonmarkman/2026/08/24/nvidia-pays-poolside-6b-to-license-its-model-factory-and-109-workers/); Yahoo Finance reports Poolside's ~$12B pre-money valuation (https://finance.yahoo.com/technology/ai/articles/nvidia-pays-6-billion-license-115520933.html); The Next Web notes the deal is expressly framed as "not an acquisition" (https://thenextweb.com/news/nvidia-poolside-6bn-model-factory-licence).
• Significance: A landmark "license-plus-talent" structure that gives Nvidia access to commercial model-training software and a key team while letting Poolside (co-founders Eiso Kant and Jason Warner) remain independent — a pattern Nvidia previously used with Groq and Enfabrica to avoid full-merger antitrust review.

…(truncated — the summary above captures the substance)

Round 0 · Finding 4

Significant AI Research & Open-Source Developments, 2026-08-20 → 2026-08-26

1. Executive Summary

The week of August 20–26, 2026 was dominated by two open-ecosystem events at the model level — an anonymous "stealth" frontier model (OX Alpha) appearing free on OpenRouter (Aug 20) and Alibaba Qwen's first open-weight preview of the Qwen4 architecture (Qwen3.8-Flash-Next, Aug 26) — plus a cluster of notable arXiv research on agentic memory/reinforcement learning and a major open video dataset release (both Aug 25). No single frontier-lab flagship launch (OpenAI/Anthropic/Google/Meta/DeepSeek) fell inside the window per the release trackers reviewed; the window's center of gravity was open-weight releases and agentic/research papers. In-window primary sources were verified by direct fetch; several headline benchmark claims (OX Alpha's coding scores) are community-reported and remain unverified against an official source.

2. Key Findings (dated, with confidence)

A. OX Alpha — anonymous frontier-class reasoning model appears on OpenRouter — 2026-08-20 (HIGH confidence on existence/specs; MEDIUM on performance claims)

OpenRouter's model page (fetched live) confirms a stealth-listed model stealth/ox-alpha, described as "a reasoning model designed for coding, sustained agentic work, and production workloads," with a 1,048,576-token context window, 131,072 max output tokens, text/image/video input, tool calling, and $0/$0 pricing during preview. The page's own FAQ states: "Ox Alpha was released on August 20, 2026", and the page banner notes it is "Going away August 26, 2026" (one-week free preview ending today). OpenRouter explicitly states the developer is "a third-party provider who has chosen to remain anonymous during this preview." https://openrouter.ai/stealth/ox-alpha

Significance/scale: Multiple independent in-window secondary reports describe immediate production adoption — e.g., explainx.ai (dated Aug 22, 2026) documents serving-layer forensics pointing to Zhipu AI's Z.AI infrastructure as the leading theory of origin (a Java stack trace), while noting Zhipu has not confirmed anything; the same article reports integration into Nous Research's Hermes Agent and the Zed editor within ~24 hours. https://www.explainx.ai/blog/ox-alpha-what-we-know-mystery-ai-model-august-2026 ; corroborated by cybersecuritynews.com and aicatchup.com coverage of the Aug 20 drop (claimed "up to 100 trillion free tokens/day" allowance — unverified). https://cybersecuritynews.com/new-mysterious-ai-model-dubbed-ox-alpha/ ; https://aicatchup.com/news/ox-alpha-stealth-model-free-openrouter-opencode

…(truncated — the summary above captures the substance)

Round 1 · Finding 1

In-Window Releases (2026-08-20 to 2026-08-26): Google DeepMind, Meta, xAI, Microsoft

Executive Summary

Of the four labs, only Google (via its Google Research / Google XR organization) and Meta had clear, date-verified in-window (2026-08-20..2026-08-26) announcements. The two "spotted" items both resolve, but with important corrections:

• AgentHands XR IS date-confirmed as an in-window release — the Google Research blog published it on August 25, 2026. Critically, it is a Google Research / Google XR project, not Meta — the prior finding misattributed it.
• WeatherNext is NOT in-window — the Google DeepMind blog is dated August 6, 2026, before the window (verifiable background, not a weekly development).

For xAI, no dated in-window release was confirmed (Grok 4.6 shipped August 12, pre-window). For Microsoft, a minor in-window roadmap post was found but no confirmed in-window model/product release. In-window sources for these four labs are thin overall; most of the week's activity at Google and Meta falls just outside the window or is month-level dated.

Key Findings

1. AgentHands XR — Date-confirmed IN-WINDOW (August 25, 2026) and attributed to Google, not Meta (High confidence)

• The Google Research blog post "AgentHands: Generating interactive hand gestures for spatially grounded agent conversations in XR" carries an explicit August 25, 2026 date, bylined to Xun Qian and Ruofei Du of Google XR (https://research.google/blog/agenthands-generating-interactive-hand-gestures-for-spatially-grounded-agent-conversations-in-xr/).
• It reports AgentHands, "an LLM-powered XR prototype that augments conversational agents with synchronized, expressive hand gestures," published at CHI 2026 and tied to Google's Android XR work. This is a research publication, not a shipped consumer product.
• The Google Research blog listing confirms August 25, 2026 as its posting date (https://research.google/blog/). (The paper itself was previewed/presented at CHI '26, April 13-17, 2026 — see the ACM DOI — so the research predates the in-window blog; the in-window item is the official Google Research blog announcement.)

2. WeatherNext (Google DeepMind) — NOT in-window; dated August 6, 2026 (High confidence)

• The Google DeepMind blog "WeatherNext: AI model achieves breakthrough in forecasting cyclones" is explicitly dated August 6, 2026 (https://deepmind.google/blog/weathernext-ai-model-achieves-breakthrough-in-forecasting-cyclones/). It announces open-sourcing WeatherNext 2, WeatherNext Cyclones, and WeatherNext 2-mini alongside a Nature paper (https://www.nature.com/articles/s41586-026-10953-2). This lies outside the Aug 20-26 window and is background for this week.

…(truncated — the summary above captures the substance)

Round 1 · Finding 2

Nvidia Q2 FY2027 Earnings Figures & Market Reaction — Research Findings

Executive Summary

The Nvidia Q2 FY2027 earnings release date is confirmed within the window (August 26, 2026) from primary official sources. However, the actual financial figures (total revenue, net income, data center revenue, forward guidance) and the post-release market reaction could NOT be verified from any dated in-window source at the time of research. The release went out after US market close on 8/26/2026, and neither official nor credible secondary pages with the actual results had been indexed yet. I do not have the numbers and will not invent them.

Key Findings

1. CONFIRMED — Release date and event details (primary source, in-window).

• Nvidia's investor-relations event calendar lists the "NVIDIA 2nd Quarter FY27 Financial Results" event dated August 26, 2026, at 2:00 PM PT (5:00 PM ET), confirming the results were officially released in-window.
  • https://investor.nvidia.com/events-and-presentations/events-and-presentations/event-details/2026/NVIDIA-2nd-Quarter-FY27-Financial-Results/default.aspx
• Nvidia's official newsroom confirmed it held a conference call on Wednesday, August 26, 2026, 2 p.m. PT (5 p.m. ET) to discuss financial results for the second quarter of fiscal year 2027, which ended July 26, 2026.
  • https://nvidianews.nvidia.com/news/nvidia-sets-conference-call-for-second-quarter-financial-results-6927195
• The Nvidia newsroom homepage (fetched 8/26/2026) shows in-window company news dated Aug 20–25, 2026 (RTX Spark/Gamescom, Jetson Orin Nano 2, Vera Rubin NVL72, Groq 3 LPX, SpaceXAI) — but the earnings press release page itself was not indexed/retrievable with figures.
  • https://nvidianews.nvidia.com/news

2. NOT VERIFIED — Actual Q2 FY2027 figures (total revenue, net income, data center revenue, forward guidance).

• I could not source the actual reported numbers for Q2 FY2027. Every search for "NVIDIA Announces Financial Results for Second Quarter Fiscal 2027" returned either the prior-year Q2 FY2026 release (Aug 27, 2025: $46.7B revenue, $41.1B data center, up 56% YoY) or pre-earnings preview/expectation articles. No indexed page carried the actual Q2 FY2027 results within the window at research time.
  • Prior-year reference (background only, Q2 FY2026): https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2026

…(truncated — the summary above captures the substance)

Round 1 · Finding 3

Qwen3.8-Flash-Next / Qwen4 — Official Release Verified (2026-08-26)

Bottom line: The "leaked" specs were real. Alibaba's Qwen team officially released Qwen3.8-Flash-Next on 2026-08-26 (in-window), and its official blog, GitHub README, Hugging Face model card, and model config.json confirm every disputed spec: 125B main-model parameters, 6B active per token, 51B N-gram embeddings, and a GDN + QSA hybrid attention architecture. The model is explicitly positioned as an early preview of the Qwen4 architecture — Qwen4 itself has NOT been released (that is a future family; only the architecture preview shipped this week). The secondary tracker's benchmark numbers are verbatim copies of Alibaba's own release figures, so there is no independent benchmark verification yet.

1. In-window release timeline (all dated 2026-08-26)

• Qwen official blog, "Qwen3.8-Flash-Next: A New Architecture, Towards Ultimate Cost-Efficiency," dated 2026/08/26, announces the open-weight release and the Qwen4-preview framing (https://qwen.ai/blog?id=qwen3.8-flash-next).
• GitHub repo QwenLM/Qwen3.8-Flash-Next — commits dated Aug 26, 2026 ("Initial commit" + "Update README.md (#1)"), README news line: "2026-08-26: We release Qwen3.8-Flash-Next" (https://github.com/QwenLM/Qwen3.8-Flash-Next/).
• Hugging Face model card Qwen/Qwen3.8-Flash-Next — live; HF API reports lastModified: 2026-08-26T12:29:54.000Z, 3,073 likes, 2,551 downloads at fetch time, repo first created 2026-08-24T08:24:59Z (the Aug 24 creation is background — the public "upcoming release" placeholder). License: qwen-community-1.0 (https://huggingface.co/Qwen/Qwen3.8-Flash-Next; https://huggingface.co/api/models/Qwen/Qwen3.8-Flash-Next).
• TechNode, Aug 26, 2026 (pre-release coverage, same day): Qwen scheduled the open-source drop "at 11 p.m. Beijing time on Aug. 26," with weights plus an FP8 version (https://technode.com/2026/08/26/alibabas-qwen-to-open-source-qwen3-8-flash-next-previewing-qwen4-architecture/).

2. Official parameter counts (from the primary sources)

• 125B-parameter main model + additional 51B N-gram embeddings + 6B parameters activated per token — stated verbatim in the blog and the GitHub README (https://qwen.ai/blog?id=qwen3.8-flash-next; https://github.com/QwenLM/Qwen3.8-Flash-Next/).
• On-disk total ≈ 180B parameters: HF API safetensors block reports BF16 total of 179,999,981,424 params (+35 int64), 131 safetensors shards, ~360 GB storage (https://huggingface.co/api/models/Qwen/Qwen3.8-Flash-Next). The difference from 125B+51B=176B is the vision tower and other components.
• Context: natively 262,144 tokens, extensible to 1,000,000 with YaRN (blog; config max_position_embeddings: 262144).
• Production sibling: the hosted "Qwen3.8-Flash" (1M default context, built-in tools) is priced at $0.16/M input and $0.47/M output tokens on QwenCloud, "API coming soon" (blog + model card).

3. Official architecture details

…(truncated — the summary above captures the substance)

Round 1 · Finding 4

OX Alpha Stealth Model — Verified Findings (2026-08-20 to 2026-08-26)

1. Executive Summary

The "OX Alpha" mystery is resolved as of 2026-08-26: Beijing-based Z.AI (formerly Zhipu AI) confirmed to Bloomberg on Wednesday 2026-08-26 that Ox Alpha is a new iteration of its GLM model series, and said it would release the model's open weights that same night. The Zhipu/Z.AI connection — initially pure forensics — is now company-confirmed via Bloomberg reporting, though OpenRouter's own listing still describes the provider as anonymous. On benchmarks, the viral "80% on DeepSWE" figure traces to a 10-task informal sample; fuller community runs of all 113 DeepSWE tasks produced lower numbers (roughly 58–63%) with conflicting reports. On availability, OpenRouter's primary record states the model is "Going away August 26, 2026", was still free ($0/$0) as of 2026-08-26, and carries an API-level expiration_date of 2026-08-26 — so the free-preview end date is confirmed, while what happens immediately after (paid vs. offline) cannot yet be observed.

2. Key Findings (with confidence levels)

3. Detailed Analysis

3.1 Provider identity: Zhipu/Z.AI — confirmed on the record (2026-08-26)

…(truncated — the summary above captures the substance)

Round 2 · Finding 1

Did Microsoft ship the Copilot super-app merge, the MAI app, or a new MAI/Copilot model between 2026-08-20 and 2026-08-26?

Headline finding: Based on the official Microsoft channels I could retrieve, I found NO evidence of an in-window (2026-08-20 to 2026-08-26) official announcement that the "Copilot super-app merge," a standalone "MAI app," or a new MAI/Copilot model shipped in that window. The pre-window rollouts I verified moved before this window, and the in-window official announcements I found are partner/ecosystem items, not core product ships. (Confidence in the negative: Medium — see Gaps.)

Key findings

1. The Copilot "merge" / unified-app rollout and the app changes all pre-date the window (before 2026-08-20).

• Microsoft's official "Updates to Copilot and the Microsoft Copilot app" support page (https://support.microsoft.com/en-us/microsoft-365-copilot/learning/changes-microsoft-copilot-app) announced the retirement of Deep Research in the consumer Copilot app starting August 18, 2026 — before the window.
• The official "Copilot on Windows release notes — Windows Insider Program" page (https://learn.microsoft.com/en-us/windows-insider/release-notes/apps/copilot-on-windows) lists its latest entry as 8/14/2026 (version 152.0.4191.25, updating the Copilot taskbar icon / navigation / feature availability); its own "Last updated" stamp reads 2026-08-14. Nothing on this official page was updated inside 8/20–8/26, so no Copilot-on-Windows app change shipped in-window per Microsoft's own release notes.
• The unified-app merge was widely reported as rolling out the week of Aug 10–13 with Deep Research, Group Chat, and Copilot Podcasts retiring August 18, 2026 (TechCrunch, dated 8/13/2026: https://techcrunch.com/2026/08/13/microsoft-kills-off-unsuccessful-ai-features-while-merging-its-separate-copilot-apps/). That is pre-window; I treat this as contamination confirming the timing, not as an in-window ship.

2. The in-window official Microsoft announcements I found are partner/ecosystem items, none of which is the super-app merge, an MAI app, or a new MAI/Copilot model.

• From the official "August 2026 announcements — Partner Center" page (https://learn.microsoft.com/en-us/partner-center/announcements/2026-august):
  • August 20, 2026 — "Dragon Copilot Physician apps and agents" availability in Microsoft Marketplace (healthcare partner platform, not a consumer/model ship).
  • August 24, 2026 — "Guide every Copilot-in-30 trial with a personalized 30-day success plan" (the Copilot Success Planner for the M365 Copilot Business trial) and a marketing ROI partnership post.

…(truncated — the summary above captures the substance)

Round 2 · Finding 2

I investigated the four highest-value open gaps. Below are my verified findings, with explicit note of what could not be confirmed before writing.

Findings

1. NVIDIA Q2 FY2027 earnings (released today, 8/26/2026, after market close)

Confirmed (in-window event): NVIDIA's Q2 fiscal year 2027 earnings were released after market close on August 26, 2026.

• NVIDIA's investor relations page for the event is dated Aug 26, 2026: "NVIDIA 2nd Quarter FY27 Financial Results" (https://investor.nvidia.com/events-and-presentations/events-and-presentations/event-details/2026/NVIDIA-2nd-Quarter-FY27-Financial-Results/default.aspx).
• Kiplinger's live blog (LiveBlogPosting, datePublished 2026-08-25T14:15Z, coverage through 2026-08-27) states results release "after the stock market closes on Wednesday, August 26" (4:20–4:30pm ET) — https://www.kiplinger.com/investing/live/nvidia-earnings-live-updates-and-commentary-august-2026.

NOT yet verifiable in-window (critical caveat): The actual Q2 FY27 revenue, Data Center revenue, and Q3 FY27 guidance figures, the after-hours share move, and any analyst comment dated 2026-08-26 are NOT yet indexed in the sources I could reach. Every search result returned either (a) Q1 FY27 figures of $81.6B revenue / $75.2B Data Center (dated 2026-05-20), (b) Q2 FY2026 figures of $46.7B (dated 2025-08-27), or (c) pre-earnings preview projections (consensus ~$92B revenue, ~$2.08 EPS). Because the release landed hours before my final capture, I cannot report the official Q2 figures, Q3 guidance, market reaction, or analyst commentary without inventing numbers. The mission's success criteria for these specific numbers remain open/unverified — the release itself is confirmed, but its financial contents are not yet retrievable. I recommend a follow-up capture after indexing (e.g., 8/27).

2. Qwen3.8-Flash-Next / Qwen4 architecture — still unverified in-window

I could not locate an official Qwen blog post or model card dated 2026-08-20 through 2026-08-26 confirming the leaked specs (125B total / 6B active params, N-gram embeddings, QSA attention). No official primary source confirming or denying these specs was found. The claims remain secondary-sourced only. Excluded/status unverified for the official architecture details.

3. OX Alpha (Zhipu/Z.AI) — unresolved in-window

I could not verify whether zai-org published Ox Alpha / GLM-5.3 weights by 2026-08-26, nor the exact checkpoint name and license, nor the OpenRouter stealth/ox-alpha status (paid/offline/renamed). No HuggingFace zai-org repository or OpenRouter status page was fetchable in my searches. The identity of OX Alpha's provider and community benchmark scores remain unverified. Excluded, status unverified.

4. OpenAI, Anthropic, DeepSeek in-window sweep (2026-08-20 → 2026-08-26)

…(truncated — the summary above captures the substance)

Round 2 · Finding 3

Z.AI Ox Alpha Open-Weight Release & OpenRouter Status — Verified 2026-08-26

Executive Summary

As of the time of this verification (2026-08-26, ~13:18–13:20 UTC), Z.AI had NOT yet published the promised Ox Alpha / GLM-5.3 open weights on HuggingFace. The HuggingFace API shows no zai-org/GLM-5.3 and no zai-org/Ox-Alpha repository — a Z.ai org member replied only "soon" in the community discussion hours after the Bloomberg-confirmed promise to release the weights "tonight" (Aug 26). On OpenRouter, stealth/ox-alpha is still listed, still free ($0/$0), not renamed, not offline — but carrying an explicit "Going away August 26, 2026" banner, meaning the free preview is scheduled to end/be removed today. Whether the weights land later tonight (China time) remains an open, time-sensitive question; the promise has not yet been fulfilled at the point of check.

Key Findings (all dated in-window: 2026-08-20 → 2026-08-26)

1. Z.AI confirmed Ox Alpha is a GLM-series model and promised weights "tonight" — 2026-08-26

PANews, publishing 2026-08-26 09:42 UTC, reported (citing Bloomberg): Zhipu (Z.AI) confirmed that Ox Alpha — which launched anonymously — "is the next-generation model in its GLM series and plans to release the model weights tonight." It adds that Ox Alpha topped OpenRouter's usage rankings with call volume "more than twice that of DeepSeek," is designed for programming/agent tasks, supports text/image/video inputs, and is "currently free to use for one week; subsequent pricing has not yet been announced." The original Bloomberg piece is titled "China's Z.AI Made Ox Alpha Stealth Model That Rivals DeepSeek" (2026-08-26).

• https://www.panews.io/articles/01a03d73-eb97-7146-99b4-b2a550e09ec6 (published 2026-08-26T09:42:00Z)
• https://www.bloomberg.com/news/articles/2026-08-26/china-s-z-ai-made-ox-alpha-stealth-model-that-rivals-deepseek (2026-08-26; paywalled, not fetched directly)
• Corroborating same-day reports: https://www.techmeme.com/260826/p15, https://officechai.com/ai/ox-alpha-z-ai/, https://www.llms.blog/posts/zhipu-confirms-ox-alpha-stealth-model-is-glm-derivative-ahead-of-open-weight-release

2. NO Ox Alpha / GLM-5.3 weights on HuggingFace as of 2026-08-26 13:18 UTC — the release had NOT landed

Direct check of the HuggingFace API for zai-org's 50 most recently modified models (queried 2026-08-26 13:18 UTC, response header date: Wed, 26 Aug 2026 13:18:49 GMT): the newest model is zai-org/GLM-5 (last modified 2026-08-11). No GLM-5.3, no Ox Alpha, no new checkpoint exists in the zai-org organization.

• https://huggingface.co/api/models?author=zai-org&sort=lastModified&direction=-1&limit=50

…(truncated — the summary above captures the substance)

Round 2 · Finding 4

Nvidia Q2 FY27 Earnings (2026-08-26): Verification Report

1. Executive Summary

The single most important fact — that Nvidia reported its fiscal Q2 2027 results after the US market close on Wednesday, 2026-08-26 (quarter ended 2026-07-26), with a 2:00 p.m. PT / 5:00 p.m. ET analyst conference call — is confirmed by multiple independently indexed sources. However, I was unable to retrieve the actual reported figures (Q2 total revenue, data-center revenue, Q3 FY27 guidance) from the official press release or any fetched mirror page: every attempt to load the primary or high-quality secondary page was blocked (empty JS-rendered body on the NVIDIA newsroom, 404 on the guessed investor-relations and Nasdaq URLs, a 402 paywall on Investopedia, an empty body on Kiplinger, and timeouts on MarketBeat/EDGAR). Per the verification rules, I will not state any Q2 FY27 actuals or the after-hours percentage move, because no fetched page contained them. What I can state with confidence is the event timeline, the pre-announced $91B ±2% guidance baseline issued by Nvidia itself on 2026-05-20, the Q1 FY27 baseline actuals, and a same-evening after-hours price level.

2. Key Findings

3. Detailed Analysis

…(truncated — the summary above captures the substance)

Round 3 · Finding 1

Verification Sweep: OpenAI / Anthropic / DeepSeek / Google DeepMind newsrooms (2026-08-20..2026-08-26) + Microsoft MAI post dates

Executive Summary

Directly fetching the four official newsrooms resolves both open gaps. OpenAI has five in-window items (Aug 20, 24, 25) — including a major silicon milestone ("Jalapeño" inference-chip results, Aug 25). Anthropic has one in-window item ($5M wellbeing-evaluation grant program, Aug 25). Google DeepMind has one in-window item (EVE Online / Fenris Creations games-AI research partnership, Aug 21). DeepSeek's official newsroom shows zero items since 2026-04-24 — confirmed absence from the primary source, not a retrieval gap. Microsoft's two MAI model posts are both OUT of window: MAI-Cyber-1-Flash-in-MDASH was published 2026-07-27 (official Microsoft Blog JSON-LD), and MAI-Code-1.1-Flash was published 2026-08-11 (GitHub Changelog JSON-LD). Neither falls inside Aug 20–26.

Key Findings

1. OpenAI — 5 in-window items (primary source: openai.com/news/, fetched)

Out-of-window boundary case: "Offering Zero Data Retention for frontier models" is dated Aug 19, 2026 (https://openai.com/news/) — excluded, one day before the window.

2. Anthropic — 1 in-window item (primary source: anthropic.com/news, fetched)

…(truncated — the summary above captures the substance)

Round 3 · Finding 2

NVIDIA Q2 FY2027 Earnings (due 2026-08-26) — Status: NOT YET ANNOUNCED as of research time

Executive Summary

The single most important finding is that NVIDIA's Q2 FY27 results have NOT been published yet. The research run occurred at ~13:25 UTC on 2026-08-26 (≈6:25 AM Pacific), which is before NVIDIA's scheduled earnings release. The SEC EDGAR primary record confirms no earnings 8-K exists: NVIDIA's most recent 8-K is dated 2026-08-17 (items 1.01, 2.03, 7.01 — a credit/debt-related filing, not earnings). Because the release has not happened, total revenue, data center revenue, Q3 FY27 guidance, and the after-hours/next-day share move CANNOT be retrieved from any source — they do not exist yet. Any numbers circulating right now are pre-announcement expectations, not results. This resolves the "major gap" from prior rounds: the gap existed because the event was scheduled for later on 2026-08-26, not because of a retrieval failure.

Key Findings (with confidence levels)

1. No earnings 8-K filed as of research time — HIGH CONFIDENCE (primary record). EDGAR's company filing list for CIK 0001045810 (fetched directly) shows the most recent 8-K is Acc-no 0001045810-26-000069, filed 2026-08-17 ("Current report, items 1.01, 2.03, and 7.01"). The prior earnings 8-K (items 2.02/9.01) was filed 2026-05-20 (Q1 FY27) and 2025-08-27 (Q2 FY26). An earnings release would appear as a new 8-K with items 2.02/9.01 (Exhibit 99.1 press release) — it is absent. Source: https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001045810&type=8-K&dateb=&owner=include&count=40

2. Release is scheduled for later on 2026-08-26, after US market close — HIGH CONFIDENCE (secondary, dated pre-window/pre-release). NVIDIA scheduled the Q2 FY27 results call for Wednesday, August 26, 2026, at 2:00 p.m. PT (5:00 p.m. ET), covering the quarter ended July 26, 2026 (schedule announcement dated 2026-07-30: https://aitoolly.com/ai-news/article/2026-07-30-nvidia-schedules-conference-call-for-second-quarter-fiscal-year-2027-financial-results). An Aug 25, 2026 preview confirms the timing: results expected "on Thursday, 27 August 2026 at around 4.20am GMT+8" (i.e., 20:20 UTC / ~1:20 PM PT on Aug 26) — https://www.ig.com/en/news-and-trade-ideas/nvidia-q2-fy2027-earnings-preview--what-to-watch-260825. NVIDIA's standard cadence is a ~1:05 PM PT press release with the 8-K filed minutes later; the call follows at 2 PM PT. The server clock on the SEC responses (13:22–13:24 UTC) confirms the run predates the ~20:05 UTC release.

…(truncated — the summary above captures the substance)

Round 3 · Finding 3

Qwen4 / Qwen3.8-Flash-Next: Do the Primary Sources Confirm the Leaked Specs?

Bottom line: The leaked specifications are CONFIRMED by the official primary sources, with two important refinements — (1) the "125B total" figure is the main model count, not the total weight count (official total is 125B main + 51B n-gram embeddings + 4B MTP), and (2) Qwen4 itself has not been released; Qwen3.8-Flash-Next is officially an architecture preview of Qwen4. The release did land in-window, on 2026-08-26, with weights live on Hugging Face.

1. Release timing — CONFIRMED, in-window (2026-08-26)

• The official GitHub README (QwenLM/Qwen3.8-Flash-Next, commit "Update README.md (#1)" dated Aug 26, 2026) states under News: "2026-08-26: We release Qwen3.8-Flash-Next." — https://github.com/QwenLM/Qwen3.8-Flash-Next/
• The Hugging Face model registry API (fetched 2026-08-26) shows Qwen/Qwen3.8-Flash-Next created 2026-08-24T08:24:59Z (the "Upcoming release" placeholder) and last modified 2026-08-26T12:29:54Z with public weights (2,551 downloads, 3,091 likes at fetch time); the FP8 variant Qwen/Qwen3.8-Flash-Next-FP8 last modified 2026-08-26T11:55:24Z — https://huggingface.co/api/models?author=Qwen&sort=lastModified&direction=-1&limit=30
• The HF model card is a full release card (license, architecture overview, official benchmark table), not a placeholder — https://huggingface.co/Qwen/Qwen3.8-Flash-Next

The planned 23:00 Beijing-time (Aug 26) ModelScope drop reported by secondary outlets is consistent with this; I could not directly load the ModelScope page (request timed out), so the ModelScope listing itself is unverified, but the Hugging Face + GitHub primary records independently confirm the Aug 26 release.

2. Architecture — CONFIRMED, and now fully specified by primary sources

The official model card and config.json (both fetched directly) confirm every leaked architectural element and add precise numbers:

…(truncated — the summary above captures the substance)

Round 3 · Finding 4

Verification Report: Did Z.AI release GLM-5.3 / Ox Alpha open weights (2026-08-26 → 2026-08-27)?

Direct answer: As of the latest check possible in this round (≈13:26 UTC / 21:26 China Standard Time on 2026-08-26), NO — Z.AI has not yet published GLM-5.3 or Ox Alpha weights on Hugging Face, and the OpenRouter route stealth/ox-alpha is still listed (not yet removed) with its "Going away August 26, 2026" banner still displayed. The promised release was for "tonight" (night of Aug 26, China time), so the verdict could flip within hours; every check below is timestamped.

1. Hugging Face: no zai-org repo exists for GLM-5.3 or Ox Alpha (verified, 2026-08-26)

• https://huggingface.co/zai-org/GLM-5.3 returns a 404 page ("Sorry, we can't find the page you are looking for") — repo does not exist (fetched 2026-08-26 ~13:26 UTC). Same for https://huggingface.co/zai-org/Ox-Alpha (404, fetched same time).
• The Hugging Face API under the zai-org author namespace returns [] for search=ox and search=GLM-5.3 (fetched 2026-08-26 13:23–13:24 UTC), i.e., no public repo ID under Z.AI contains either string.
• The official org page (https://huggingface.co/zai-org/models, fetched 2026-08-26 ~13:25 UTC) lists 150 models across 5 pages; page 1 sorted by "Recently updated" shows the newest update as zai-org/GLM-5 "Updated 15 days ago" (~Aug 11). No GLM-5.3 or Ox-Alpha repo appears anywhere in the visible listing.
• One caveat/signal: the org "Recent Activity" feed shows zai-org/GLM-5.2:glm5.3 activity "about 4 hours ago" (≈09:25 UTC) by team member ZHANGYUXUAN-zR. This is consistent with weights being staged inside the existing GLM-5.2 repo (a pattern matching the earlier secondary claim that GLM-5.3 open weights were "staged behind safety review" — https://www.explainx.ai/blog/glm-5-3-launch-cyber-defense-benchmarks-august-2026), but I could not verify the content of that file/activity before the round ended (the GLM-5.2 repo fetch timed out). Treat as unverified staging signal, not a release.

2. OpenRouter: stealth/ox-alpha is STILL listed as of check time (verified, 2026-08-26 ~13:23 UTC)

…(truncated — the summary above captures the substance)

Investigation Trail

Round 0

• What major AI model releases, product launches, or feature announcements (e.g., new foundation models, model upgrades, AI hardware/devices) were made by companies like OpenAI, Google/DeepMind, Anthropic, Meta, xAI, Mistral, Microsoft, Nvidia, or others between 2026-08-20 and 2026-08-26?
• What significant corporate and business developments involving AI companies occurred between 2026-08-20 and 2026-08-26 — for example, major funding rounds, acquisitions, partnerships, IPO/financial results, or leadership changes at leading AI labs or AI infrastructure companies?
• What AI policy, regulatory, safety, or legal developments were announced or occurred between 2026-08-20 and 2026-08-26 — e.g., new laws or executive orders, court rulings, EU AI Act / GDPR actions, US federal or state agency decisions, AI safety evaluations, or major AI-related incidents?
• What notable AI research breakthroughs, open-source releases, or benchmark results were published between 2026-08-20 and 2026-08-26 that researchers or practitioners widely consider significant — e.g., advances in reasoning, multimodal systems, agentic AI, training efficiency, or interpretability/safety?

Round 1

• What architecture details, parameter counts, and official benchmark results did Alibaba/Qwen publish for the Qwen3.8-Flash-Next / Qwen4 release between 2026-08-20 and 2026-08-26, and how do the official model card or blog figures reconcile with the leaked specs (125B total / 6B active parameters, N-gram embeddings, QSA attention) and secondary-tracker benchmarks?
• What were Nvidia's actual Q2 FY2027 earnings figures (total revenue, net income, data center revenue, and forward guidance) reported in its earnings release between 2026-08-20 and 2026-08-26, and what was the immediate market reaction (share price move and analyst commentary)?
• What verified information emerged between 2026-08-20 and 2026-08-26 about the identity of the 'OX Alpha' stealth model on OpenRouter — who is the actual provider (is the Zhipu/Z.AI connection confirmed or refuted), what benchmark scores did the community verify, and is the model still freely available or did it move to paid/offline after 2026-08-26?
• Which dated announcements between 2026-08-20 and 2026-08-26 can be verified from Google DeepMind, Meta, xAI, or Microsoft — specifically, are the spotted 'WeatherNext' and 'AgentHands XR' items date-confirmed as in-window releases, and what other model releases, product launches, or major research publications from these labs carry explicit dates inside that window?

Round 2

• What were Nvidia's actual Q2 FY27 results announced on 2026-08-26 (total revenue, data-center revenue, and Q3 FY27 guidance per the official press release), and what was the after-hours market reaction and analyst commentary published on 2026-08-26?
• Did Z.AI deliver the promised Ox Alpha open-weight release on 2026-08-26 — search HuggingFace under the zai-org organization for GLM-5.3/Ox Alpha repositories, confirm the exact checkpoint name and license — and what is the status of OpenRouter's stealth/ox-alpha model as of 2026-08-26 (still listed, paid, offline, renamed, or removed)?
• Which AI models, products, or research releases did OpenAI, Anthropic, and DeepSeek announce on their official newsrooms or blogs between 2026-08-20 and 2026-08-26 (inclusive), with exact publication dates and URLs?
• Did Microsoft actually ship the Copilot super-app merge, the MAI app, or any new MAI/Copilot model between 2026-08-20 and 2026-08-26, according to official Microsoft AI, Windows, and Copilot news pages or release notes (not the Dynamics 365 roadmap post)?

Round 3

• What are NVIDIA's Q2 FY27 financial results announced on 2026-08-26, pulled from the official 8-K filed with SEC EDGAR (CIK 0001045810, Exhibit 99.1) or the NVIDIA IR press release — specifically total revenue, data center segment revenue, Q3 FY27 guidance, and the immediate after-hours share move on 2026-08-26 / next-day trading move on 2026-08-27?
• Did Z.AI actually release GLM-5.3 / Ox Alpha as open weights between 2026-08-26 and 2026-08-27 — check huggingface.co/api/models?author=zai-org for a new repo, verify the model card, and confirm whether the OpenRouter route 'stealth/ox-alpha' is still listed or was removed after its 'Going away August 26, 2026' banner?
• Which announcements published on the official AI newsrooms of OpenAI, Anthropic, DeepSeek, and Google DeepMind fall strictly between 2026-08-20 and 2026-08-26 (use site-restricted searches such as site:openai.com, site:anthropic.com, site:deepseek.com, site:blog.google/technology/ai), and separately, can the publish dates of Microsoft's MAI-Code-1.1-Flash and MAI-Cyber-1-Flash-in-MDASH announcements be verified — via archive.org, Bing cache, or rendered page metadata — as falling inside that same window?
• Do the official Qwen blog and Qwen model cards confirm or contradict the leaked specifications for Qwen4 and Qwen3.8-Flash-Next reported between 2026-08-20 and 2026-08-26 — what do the primary sources say about release timing, architecture, and parameter counts (e.g., the claimed 125B total / 6B active parameters for Qwen3.8-Flash-Next)?

Sources

• https://openai.com/index/jalapeno-first-results/
• https://openai.com/index/the-full-stack-behind-abundant-intelligence/
• https://openai.com/index/gpt-5-6-in-kiro/
• https://openai.com/news/
• https://openai.com/index/introducing-ai-futures/
• https://aireleasetracker.com/model/qwen/qwen3.8-flash-next
• https://support.claude.com/en/articles/12138966-release-notes
• https://aireleasetracker.com/latest
• https://llmgateway.io/timeline
• https://www.digitalapplied.com/blog/ai-model-releases-august-2026-tracker
• https://benchlm.ai/model-updates/releases/august-2026
• https://vfuturemedia.com/ai/new-ai-models-released-august-2026/
• https://llmgateway.io/timeline/2026
• https://lmmarketcap.com/llm-updates
• https://felloai.com/best-ai-models/
• https://skycrumbs.com/blog/ai-models-august-2026
• https://capitalandcompute.net/blog/new-ai-models-august-2026/
• https://openai.com/newsroom/
• https://openai.com/news/product-releases/
• https://releasebot.io/updates/openai
• https://decodingdatascience.com/openai-august2026/
• https://www.buildfastwithai.com/blogs/collection/ai-industry-news-trends
• https://pricepertoken.com/news/model-releases
• https://aibriefing.dev/
• https://llm-stats.com/llm-updates
• https://cdn.openai.com/pdf/GPT_5_6_August_Updates.pdf
• https://deepmind.google/blog/
• https://deepmind.google/
• https://skycrumbs.com/blog/google-deepmind-august-2026
• https://blog.google/innovation-and-ai/technology/ai/google-io-2026-all-our-announcements/
• https://blog.google/innovation-and-ai/technology/developers-tools/google-io-2026-collection/
• https://cybernews.com/ai-news/google-io-2026-gemini-omni-antigravity-agentic-ai/
• https://io.google/2026
• https://research.google/blog/
• https://code.claude.com/docs/en/whats-new/2026-w32
• https://releasebot.io/updates/anthropic/claude
• https://releasebot.io/updates/anthropic
• https://adam.holter.com/every-new-claude-launch-since-january-2026-full-timeline/
• https://techdailyshot.com/blog/anthropic-claude-workflow-automation-update-august-2026
• https://blog.mean.ceo/anthropic-claude-news-august-2026/
• https://claude.com/blog-category/announcements
• https://aitoolsrecap.com/News/claude
• https://www.anthropic.com/news
• https://www.transparencycoalition.ai/news/ai-legislative-update-august21-2026
• https://www.promptinjection.net/p/ai-llm-news-roundup-august-09-august-22-2026
• https://techcrunch.com/2026/08/22/openai-says-california-should-strengthen-its-ai-safety-bill/
• https://techcrunch.com/2026/08/22/frontier-ai-labs-still-wont-say-how-theyd-contain-a-rogue-model/
• https://www.reuters.com/technology/artificial-intelligence/ai-firms-cant-yet-contain-what-theyve-built-study-finds-2026-08-19/
• https://www.reuters.com/world/americas/brazil-launches-ai-supercomputer-push-splits-projects-between-chinese-us-firms-2026-08-20/
• https://www.reuters.com/world/china/china-urges-respect-digital-sovereignty-ai-race-2026-08-19/
• https://www.nature.com/articles/s41746-026-03165-5
• https://commission.europa.eu/news-and-media/news/safer-and-more-transparent-ai-2026-08-02_en
• https://eur-lex.europa.eu/eli/reg/2026/1744/oj/eng
• https://www.whitehouse.gov/presidential-actions/2026/06/promoting-advanced-artificial-intelligence-innovation-and-security/
• https://af.net/realtime/ai-regulation-news-august-2026-the-enforcement-era-begins-us-gridlock-ongoing/
• https://cubbbix.com/blog/ai-regulation-august-2026-global-update/
• https://www.regulation-ai.eu/en/blog/eu-ai-act-enforcement-august-2026/
• https://www.aiandnews.com/blog/ai-regulations-august-2026/
• https://www.technology.org/2026/07/17/eu-ai-act-what-actually-applies-on-2-august-2026/
• https://www.themodernblog.com/ai-regulation-news/
• https://www.digitalapplied.com/blog/eu-ai-act-august-2026-transparency-obligations-agency-checklist
• https://learn.oreateai.com/learn/how-august-2026-reshaped-the-global-landscape-of-ai-safety-regulation
• https://www.artificialintelligence-news.com/categories/inside-ai/new_governance-regulation-and-policy/
• https://af.net/realtime/how-august-2026-reshaped-the-global-landscape-of-ai-safety-regulation/
• https://www.ismscopilot.com/learn/eu-ai-act-what-applies-from-2-august-2026
• https://www.european.express/2026/07/17/eu-ai-act-what-actually-applies-on-2-august-2026/
• https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act
• https://fpf.org/wp-content/uploads/2026/07/FPF-EU-AI-Act-Timeline-2026-R2.pdf
• https://aiadopt.eu/en/insights/august-2026-ai-act-enforcement
• https://perspectivelabs.org/eu-ai-act-enforcement-august-2026/
• https://itknowledgelab.com/blog/eu-ai-act-enforcement-august-2026-compliance
• https://www.hklaw.com/en/insights/publications/2026/04/us-companies-face-eu-ai-acts-possible-august-2026-compliance-deadline
• https://informedclearly.com/en/ai/55795/eu-ai-act-compliance-deadline-2026
• https://informedclearly.com/en/ai/60895/eu-ai-act-enforcement-deadlines-2026
• https://vorplabs.com/ai-regulatory-updates/united-states
• https://aitoolly.com/ai-news/2026-08-25
• https://www.aiapps.com/blog/ai-news-august-breakthroughs-launches-trends-cant-miss/
• https://www.aibase.com/daily
• https://kraviona.com/blog/latest-ai-news-august-2026
• https://llm-stats.com/ai-trends
• https://aiweekly.co/ai-news-today
• https://www.artificialintelligence-news.com/
• https://aitoolsrecap.com/Blog/AINewsAugust2026.aspx
• https://cdn.ca9.uscourts.gov/datastore/opinions/2026/08/04/26-1444.pdf
• https://arstechnica.com/tech-policy/2026/08/suspecting-court-of-using-ai-man-injected-prompts-in-filings-to-try-to-win-case/
• https://caselaw.findlaw.com/court/us-9th-circuit/294834.html
• https://www.ropesgray.com/en/sites/Artificial-Intelligence-Court-Order-Tracker
• https://axis-intelligence.com/ai-copyright-lawsuits-tracker/
• https://techcrunch.com/2026/08/01/judge-denies-xais-request-to-block-minnesota-ban-on-nudify-apps/
• https://variety.com/2026/digital/news/suno-loses-ai-lawsuit-gema-1236825010/
• https://www.gunder.com/en/news-insights/insights/2026-ai-laws-update-key-regulations-and-practical-guidance
• https://www.govinfo.gov/content/pkg/DCPD-202600376/pdf/DCPD-202600376.pdf
• https://www.congress.gov/crs_external_products/IF/PDF/IF13268/IF13268.2.pdf
• https://www.federalregister.gov/documents/2026/06/05/2026-11415/promoting-advanced-artificial-intelligence-innovation-and-security
• https://www.whitehouse.gov/presidential-actions/executive-orders/
• https://www.governmentcontractslaw.com/2026/06/ai-heats-up-new-executive-order-on-promoting-advanced-artificial-intelligence-innovation-and-security/
• https://techjournal.org/trump-ai-executive-order-cybersecurity-2026
• https://www.federalregister.gov/presidential-documents/executive-orders/donald-trump/2026
• https://tech-insider.org/trump-ai-executive-order-caisi-2026/
• https://thefederalregister.org/documents/2026-11415/promoting-advanced-artificial-intelligence-innovation-and-security
• https://thirdruntime.com/?date=2026-08-22
• https://www.buildfastwithai.com/blogs/ai-news-today-august-22-23-2026
• https://artificialintelligenceherald.com/ai-news-today
• https://techcrunch.com/2026/08/
• https://www.opensourceforu.com/2026/08/nvidia-signs-deal-with-poolside/
• https://www.forbes.com/sites/jonmarkman/2026/08/24/nvidia-pays-poolside-6b-to-license-its-model-factory-and-109-workers/
• https://finance.yahoo.com/technology/ai/articles/nvidia-pays-6-billion-license-115520933.html
• https://thenextweb.com/news/nvidia-poolside-6bn-model-factory-licence
• https://www.marketbeat.com/earnings/reports/2026-8-26-nvidia-co-stock/
• https://www.fool.com/investing/2026/08/25/nvda-stock-earnings-q2-date-aug-26/
• https://www.forbes.com/sites/paulocarvao/2026/08/25/nvidia-earnings-a-scorecard-for-the-ai-boom/
• https://www.theneuron.ai/digest/everything-that-happened-in-ai-today-friday-august-21-2026/
• https://techstartups.com/2026/08/21/top-tech-news-today-august-21-2026-anthropic-apple-broadcom-google-nvidia-openai-tesla-more/
• https://www.cnbc.com/2026/02/27/open-ai-funding-round-amazon.html
• https://www.nytimes.com/2026/02/27/business/openai-funding.html
• https://www.cnbc.com/2026/03/31/openai-funding-round-ipo.html
• https://www.anthropic.com/news/series-h
• https://fortune.com/2026/08/15/anthropic-revenue-q2-11-5-billion-ipo-investors/
• https://techcrunch.com/2026/08/12/ai-coding-startup-cognition-reportedly-already-in-talks-to-raise-at-40b-valuation/
• https://x.ai/news/series-e
• https://techstartups.com/2026/08/04/venture-capital-startup-funding-roundup-august-4-2026-a16z-base10-dream-ventures-jpmorgan-sequoia-capital-valor-equity-partners-yc-more/
• https://techstartups.com/2026/08/05/venture-capital-startup-funding-roundup-august-5-2026-a16z-blackrock-base10-franklin-templeton-general-catalyst-insight-partners-y-combinator-more/
• https://skycrumbs.com/blog/ai-startup-funding-august-2026
• https://aifunding.me/deals
• https://fundup.ai/recently-funded-startups
• https://www.joineta.org/blog/ai-investment-roundup-august-2026
• https://www.secondtalent.com/resources/ai-startup-funding-investment/
• https://adtools.org/buyers-guide/ai-news-openai-110-billion-funding-round
• https://the-agent-report.com/2026/08/ai-agent-funding-surge-august-2026/
• https://roundly.io/top-50
• https://www.ai-market-watch.com/news/category/acquisition
• https://justainews.com/category/companies/acquisitions/
• https://ir.dynatrace.com/news-events/press-releases/detail/435/dynatrace-to-acquire-ai-observability-leader-arize
• https://www.dynatrace.com/news/press-release/dynatrace-to-acquire-arize/
• https://www.cnbc.com/2026/08/06/amd-buys-taalas-startup-that-hardwires-ai-models-into-its-silicon.html?msockid=1ae1f36717426da41419e4a616606c65
• https://presenc.ai/research/ai-acquisition-tracker-2026
• https://americanbazaaronline.com/2026/08/20/from-energy-to-ai-five-major-us-business-deals-of-2026-486728/
• https://pricepertoken.com/news/funding
• https://af.net/realtime/latest-trends-in-ai-acquisitions-and-ma-for-august-2026/
• https://justainews.com/category/companies/funding-news/
• https://tech-insider.org/openai-110-billion-funding-round-2026/
• https://techcrunch.com/2026/02/27/openai-raises-110b-in-one-of-the-largest-private-funding-rounds-in-history/
• https://www.foxbusiness.com/technology/openais-110b-funding-round-draws-investment-from-amazon-nvidia-softbank
• https://www.forbes.com/sites/renanaashkenazi/2026/03/02/openais-110-billion-reveals-how-ai-now-finances-itself/
• https://adtools.org/buyers-guide/ai-news-openai-110b-funding-round
• https://openai.com/index/scaling-ai-for-everyone/
• https://www.pymnts.com/news/artificial-intelligence/2026/cognition-ai-eyes-40-billion-valuation-from-new-funding/
• https://www.masternodeai.com/en/news/cognition-40b-valuation-talks-1b-arr-devin
• https://creati.ai/ai-news/2026-08-14/cognition-reportedly-seeks-40b-valuation-just-months-after-devin-funding-round/
• https://valueaddvc.com/pulse/cognition-ai-40-billion-valuation-funding-talks-2026
• https://aitoolly.com/ai-news/article/2026-08-13-us-ai-startup-cognition-seeks-1-billion-funding-round-at-a-massive-40-billion-valuation
• https://www.bloomberg.com/news/articles/2026-08-12/ai-startup-cognition-in-new-funding-talks-at-40-billion-value
• https://insight.tmcnet.com/insight/cognition-ai-enters-funding-talks-at-40-billion-plus-valuation-msqnguko
• https://aventure.vc/news/2026-08-12-ai-coding-startup-cognition-reportedly-already-in-talks-to-raise-at-40b-valuation
• https://valueaddvc.com/pulse/cognition-devin-40-billion-valuation-talks-2026
• https://tracxn.com/d/companies/anthropic/__SzoxXDMin-NK5tKB7ks8yHr6S9Mz68pjVCzFEcGFZ08/funding-and-investors
• https://tech-insider.org/anthropic-65-billion-series-h-965-billion-valuation-2026/
• https://valueaddvc.com/company/anthropic
• https://valueaddvc.com/blog/anthropic-valuation-2026-965b-series-h-and-the-ipo-filing-explained
• https://www.cbinsights.com/company/anthropic/financials
• https://presenc.ai/research/ai-lab-funding-leaderboard-2026
• https://apito.ai/en/blog/news/anthropic-h-round-65b-funding-2026/
• https://getlatka.com/companies/anthropic
• https://finance.yahoo.com/calendar/earnings/
• https://www.kiplinger.com/investing/live/nvidia-earnings-live-updates-and-commentary-august-2026
• https://finquota.com/earning-calendar/
• https://zenvesto.com/earnings-calendar
• https://marketwise.com/investing/big-tech-ai-earnings-alphabet-amazon-microsoft-meta-apple-tesla/
• https://earningslabs.com/calendar/2026/08
• https://fiscal.ai/earnings-calendar/
• https://askthedate.com/UTC/en/earnings-calendar/month/
• https://tech-insider.org/nvidia-earnings-81-billion-quarter-2026/
• https://www.marketbeat.com/stocks/NASDAQ/NVDA/earnings/
• https://investor.nvidia.com/news/press-release-details/2025/NVIDIA-Announces-Financial-Results-for-Second-Quarter-Fiscal-2026/default.aspx
• https://www.sec.gov/Archives/edgar/data/1045810/000104581025000207/q2fy26pr.htm
• https://www.zacks.com/stock/research/NVDA/earnings-calendar
• https://quarter-results.com/financial-reports/nasdaq/nvda/q2/2026/nvda-q2-2026-earnings-report/
• https://tracxn.com/d/companies/xai/__saKrxbHN3TRWW-I4lYH6zkx6N5P_kMTqlLcKTzWs2ug/funding-and-investors
• https://www.techstackipo.com/h1-2026-funding-mega-rounds
• https://adtools.org/buyers-guide/ai-news-xai-20b-series-e-funding
• https://valueaddvc.com/company/xai
• https://tracxn.com/d/companies/xai/__saKrxbHN3TRWW-I4lYH6zkx6N5P_kMTqlLcKTzWs2ug
• https://mungomash.com/orgs/xai/financials/
• https://exa.ai/websets/directory/xai-funding
• https://aitoolly.com/ai-news/2026-08-21
• https://manaknightdigital.com/blog/ai-news-week-of-august-21-2026
• https://aisdg.substack.com/p/ai-news-for-august-21-2026
• https://www.marketingprofs.com/opinions/2026/55655/ai-update-august-21-2026-ai-news-and-views-from-the-past-two-weeks
• https://www.aiapps.com/blog/august-2026-ai-mega-update-major-breakthroughs-launches/
• https://www.gurufocus.com/news/9047489/nvidia-nvda-invests-10-billion-in-poolsides-ai-development
• https://enterprisedna.co/resources/news/nvidia-poolside-6-billion-model-factory-license-enterprise-coding-2026/
• https://andrew.ooo/answers/what-is-nvidia-poolside-model-factory-deal-august-2026/
• https://www.pymnts.com/news/artificial-intelligence/2026/nvidia-pays-6-billion-to-license-poolside-ai-model-development-software/
• https://cryptobriefing.com/nvidia-6b-poolside-ai-licensing-deal/
• https://finance.yahoo.com/technology/ai/articles/nvidia-pay-poolside-6-billion-181448803.html
• https://www.ai-market-watch.com/news/category/funding
• https://www.fundedstartupsdaily.com/raises/ai-ml/
• https://www.crescendo.ai/news/latest-vc-investment-deals-in-ai-startups
• https://aifundingtracker.com/ai-startup-funding-news-today/
• https://openrouter.ai/stealth/ox-alpha

Trace Index

Tool-call traces are persisted under /srv/swarm_web_runs/run-1787749863471-0004/traces.