# Major AI labs: October 6, 2026 coverage audit and source ledger

Research cutoff and source access date: **2026-10-06**.

This is an editorial selection of twelve major model developers and AI platform organizations: OpenAI, Anthropic, Google DeepMind, Meta, DeepSeek, Alibaba Qwen, xAI / SpaceXAI, Mistral, Microsoft AI, Amazon, NVIDIA, and Cohere. It emphasizes September 1–October 6 announcements with implications for agent engineering. It is not a census of every AI lab or every announcement in that period. Amazon's selected item is a jointly developed platform preview, not an Amazon-authored model release.

## Method and evidence boundaries

The audit inspected all 138 existing local blog articles, the category registry, relevant reference types, foundation and implementation docs, and the live blog index. The live `/blog` returned HTTP 200 on October 6; its 139 article-shaped links consisted of the 138 local article slugs plus the `start-here` navigation route. This verifies index agreement at the audit time, not a line-by-line comparison of deployed article bodies.

Searches covered lab names, model and product names, and the specific candidate release identifiers. Existing general discussion of a lab or managed agents was not treated as a dedicated assessment of a later release. Existing related articles are identified below to make the incremental scope inspectable.

Official announcement pages, developer documentation, repository documentation, and a versioned research preprint support factual claims. Articles distinguish those facts from original analysis and proposed experiments. No paid model evaluation, account-specific product test, cloud deployment, hardware test, training run, or independent benchmark reproduction was performed. Repository documentation inspection is not a source-code security audit. AI assisted the research and writing; no independent human review is claimed.

Source dates below are publication or explicitly dated release-entry dates. Living documentation is labeled as such; the access date is not substituted for a launch date. Announced, preview, restricted-access, and available states remain separate. The inline citations in each article bind claims to their sources.

## Coverage decisions

| Lab | Existing coverage inspected | New assessment and distinct contribution | Series action |
|---|---|---|---|
| OpenAI | [Astra launch](/blog/welcome-gpt-6-astra-research-analysis), [full-stack map](/blog/openai-2026-full-stack-research-map), [managed harnesses](/blog/managed-agent-harness-control-boundaries) | [Post-DevDay economics and persistent agents](/blog/openai-devday-2026-sol-dots-agent-economics): September 29 Sol, Dots, computer use, and continuing-work evaluation | Reuse `openai-research` |
| Anthropic | [Claude family](/blog/claude-5-model-family-fable-mythos-opus-sonnet), [platform](/blog/anthropic-claude-platform-managed-agents-api-2026) | [Claude 5.5 assessment](/blog/anthropic-claude-5-5-efficiency-safeguards-assessment): September 22/28 releases, fallback attribution and accepted-work economics | Reuse `anthropic-research` |
| Google DeepMind | Cross-lab research articles contain Google references; no dedicated Google series in the audited registry | [Argon and private memory](/blog/google-gemini-4-argon-private-memory-assessment): restricted access, output budget, confidentiality versus memory correctness | Create `google-deepmind-research` |
| Meta | [September 26 stack](/blog/meta-muse-agent-stack-spark-code-glimmer), [security audit](/blog/meta-muse-harness-architecture-security-review), [research roadmap](/blog/personal-agent-harness-research-roadmap-after-muse) | [Small Business expansion](/blog/meta-muse-small-business-approval-economics): September 29 connector and skills expansion, review burden and business approvals | Reuse `meta-research` |
| DeepSeek | [August harness audit](/blog/deepseek-harness-architecture-security-review) | [V4.1 Flash](/blog/deepseek-v4-1-flash-cache-migration-assessment): model/cache changes and conflicting migration notices | Create `deepseek-research`; move the existing harness article's category assignment here without changing its URL or body |
| Alibaba Qwen | Cross-lab harness research; no dedicated Qwen series in the audited registry | [Planner-Agent paper](/blog/qwen-planner-agent-co-evolution-evidence): September 24 v1, environment fidelity and development/test separation | Create `qwen-research` |
| xAI / SpaceXAI | General cross-lab references; no dedicated lab series | [Grok 4.7 and Team Bots](/blog/xai-grok-4-7-team-bots-shared-memory): September 21/28, native harness training and shared/private context | Create `xai-research` |
| Mistral | General cross-lab references; no dedicated lab series | [Vibe knowledge and artifacts](/blog/mistral-vibe-knowledge-base-artifact-assessment): September 22 entries, phased migration, memory correction and snapshot semantics | Create `mistral-research` |
| Microsoft AI | Microsoft agent-framework and platform references; no MAI lab series | [October 1 audio models](/blog/microsoft-mai-streaming-voice-action-boundaries): partial transcripts, speculative work and irreversible actions | Create `microsoft-ai-research` |
| Amazon | Bedrock and Nova references in broader essays; no dedicated Amazon series | [Bedrock Managed Agents](/blog/amazon-bedrock-managed-agents-aws-boundaries): September 29 preview, execution placement and identity handoff | Create `amazon-research` |
| NVIDIA | Infrastructure and research references; no dedicated NVIDIA series | [OpenShell and Sentry](/blog/nvidia-openshell-sentry-agent-safety-assessment): September 28 software/reference-design split and containment evidence | Create `nvidia-research` |
| Cohere | No dedicated Cohere series in the audited registry | [Embed 5 and translation](/blog/cohere-embed-5-retrieval-translation-assessment): September 30/9 releases, index migration and multilingual provenance | Create `cohere-research` |

## Primary source ledger

All sources below were accessed October 6, 2026. “Supports” describes the limited factual use, not an endorsement of every vendor claim.

### OpenAI

- [GPT-6.1 Sol](https://openai.com/index/introducing-gpt-6-1-sol/) — September 29, 2026; first-party model announcement. Supports pricing, reported task profile, and factuality sample limitations. No independent reproduction.
- [DevDay 2026 recap](https://openai.com/index/devday-2026-recap/) — September 29, 2026; product announcement. Supports Dots, Agents API computer use, event automations, and differentiated rollout states. The recap is not an API execution contract.
- [API changelog](https://developers.openai.com/api/docs/changelog) — living documentation; September 29 entries used. Supports the date of computer-use support and Sol API release.

### Anthropic

- [Opus 5.5](https://www.anthropic.com/claude-opus-5-5) — September 22, 2026; first-party model announcement. Supports price-versus-workload-cost distinction and fallback methodology. Safety results and productivity examples remain publisher reports.
- [Sonnet 5.5](https://www.anthropic.com/claude-sonnet-5-5) — September 28, 2026; first-party model announcement. Supports positioning and reported speed, cost, and benchmark comparisons. No cross-vendor ranking is inferred.

### Google DeepMind

- [Gemini 4 Argon](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/) — September 30, 2026; restricted-release announcement. Supports Fairwind access, expanded output limit, selected vendor results, and internal review descriptions. General availability is not assumed.
- [Private AI Compute server-side memory](https://deepmind.google/blog/advancing-private-ai-compute-with-secure-server-side-memory/) — September 23, 2026; architecture announcement. Supports the proposed encrypted persistent-memory design. No deployment or cryptographic audit was performed; no integration with Argon is inferred.

### Meta

- [Muse for Small Business](https://about.fb.com/news/2026/09/introducing-muse-small-business/) — September 29, 2026; product announcement. Supports new business connectors and skills, the stated approval promise, and initial geography. Promise not independently tested.
- [Introducing Muse](https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/) — September 8, 2026, page later updated; historical architecture context only. The new article does not re-review the earlier launch.

### DeepSeek

- [V4.1 Flash launch](https://deepseek.com/en/news/deepseek-v4-1-flash/) — September 10, 2026; model announcement. Supports architecture and cache claims. Its V4 Pro redirection notice conflicts with the API continuation notice below.
- [API changelog](https://api-docs.deepseek.com/updates/) — living documentation; September 10 entry used. Supports Flash alias routing and says V4 Pro will continue after September 14 with billing unchanged. The article preserves the public discrepancy and requires current operational verification rather than declaring retirement.

### Alibaba Qwen

- [Qwen-Planner-Agent v1 record](https://arxiv.org/abs/2609.29892v1) and [full paper](https://arxiv.org/html/2609.29892v1) — September 24, 2026; research preprint by Alibaba's MAI team. Supports the development loop, hybrid environments, CARE, fixed-checkpoint ablations, and the explicit development-set status of evolve feedback. A preprint is not peer-review certification; no results were reproduced. The separate project landing page was not accessible during this review, so installability is not asserted.

### xAI / SpaceXAI

- [Grok 4.7](https://x.ai/news/grok-4-7) — September 21, 2026; model announcement. Supports the reported training changes and native harness familiarity. Portability remains an evaluation question.
- [Team Bots](https://x.ai/news/team-bots) — September 28, 2026; public-beta announcement. Supports shared skills, separate personal contexts, credentials, and internal examples. Selected case studies do not establish causal productivity gains or tenant-specific isolation guarantees.

### Mistral

- [Release notes](https://docs.mistral.ai/resources/release-notes) — living documentation; September 22, 2026 entries used. Supports Knowledge Base preview, artifact features, and migration dates. Entry date and rollout start are distinguished.
- [Mini apps](https://docs.mistral.ai/vibe/work/mini-apps) — living product documentation. Supports snapshot and sharing behavior. No generated app or workbook was executed.

### Microsoft AI

- [Streaming transcription and voice launch](https://microsoft.ai/news/our-first-streaming-transcription-model/) — October 1, 2026; first-party model announcement. Supports partial/final transcript behavior, language counts, and availability distinctions. All timing claims are vendor reports.
- [Code of Conduct consultation](https://microsoft.ai/news/mai-code-of-conduct/) — September 14, 2026; draft policy announcement. Used only to distinguish stated objectives from measured enforcement.

### Amazon

- [Bedrock Managed Agents](https://aws.amazon.com/bedrock/managed-agents-openai/) — living product page; preview associated with September 29 announcement. Supports the AWS/OpenAI/AgentCore combination and AWS location claim.
- [AgentCore getting-started guide](https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/runtime-get-started-bma.html) — living preview documentation. Supports managed-harness versus command-environment placement, session microVMs, roles, and finite lifetime. Tutorial not executed.
- [OpenAI DevDay recap](https://openai.com/index/devday-2026-recap/) — September 29, 2026; corroborates the joint announcement date. It does not imply a new Amazon model.

### NVIDIA

- [Open Agent Safety Platform](https://nvidianews.nvidia.com/news/open-agent-safety-platform) — September 28, 2026; platform announcement. Supports the OpenShell software / Sentry reference-design distinction and reported containment behavior. Hardware performance not measured.
- [OpenShell repository](https://github.com/NVIDIA/OpenShell) — living repository documentation, checked October 6. Supports the documented enforcement and policy-checking approach. No commit-pinned code audit or proof verification is claimed.
- [NVIDIA and Palantir supply-chain collaboration](https://nvidianews.nvidia.com/news/nvidia-and-palantir-bring-sovereign-intelligence-to-critical-supply-chains) — September 10, 2026; first-party deployment announcement. Used as a domain-control example, not controlled ROI evidence.

### Cohere

- [Release notes](https://docs.cohere.com/v1/changelog) — living documentation; September 30 Embed 5 and September 9 North Small Translate entries used. Supports release dates and shared Pro/Fast space.
- [Embed model documentation](https://docs.cohere.com/v1/docs/cohere-embed) — living reference. Supports dimensions, modalities, and input limits. No compatibility with prior-generation indexes is inferred.
- [North Small Translate](https://docs.cohere.com/v1/docs/north-small-translate-1.0) — living reference. Supports translation scope and deployment/license distinctions. The guessed marketing-blog URL was inaccessible and is not used as evidence.

## Deliberate exclusions

The Astra launch, Muse's original launch/Connect hardware expansion, and the August DeepSeek Harness architecture are already covered and are linked as background instead of rewritten. Fundraising, hiring, partnerships without relevant technical evidence, and rumors were not used to pad a lab's article. Earlier announcements appear only where they explain the new assessment.

This batch selects a substantive recent development for each named organization. It does not claim to cover every science, safety, product, or business announcement from those organizations. Future articles can extend each series without rewriting this dated evidence snapshot.