This observation covers technical material published around the period from 07:44 on 29 October to 07:44 on 30 October 2026 in Australia/Sydney. It includes three verified changes.
Mistral Large 4 advances from API preview to open weights
Mistral released the improved Mistral Large 4 weights on 29 October, moving the model beyond the public API preview introduced on 6 October to a downloadable, self-hosted and customisable release. It is a natively multimodal mixture-of-experts model with about one trillion total parameters and 49 billion parameters active per token. It accepts text and images, produces text, and supports a one-million-token context. Official FP8 and NVFP4 variants are available, with documented paths through Transformers, vLLM, SGLang and llama.cpp. Hosted access remains available through Mistral Studio, the Mistral API and OpenRouter.
The weights use the Mistral Ecosystem Open-Weight License. It allows use, modification, fine-tuning and ordinary commercialisation, while operators of Model-as-a-Service offerings need a separate commercial licence from Mistral. It therefore carries a condition that an unmodified MIT licence would not impose. The 29 October version also adds low, medium and high reasoning modes. Mistral reports that its Terminal-Bench 4.0 result rose by 13 percentage points from the preview to 41 per cent while output tokens fell by 30 per cent. Hugging Face confirms the two official weight variants, licence and day-zero inference support. Most capability figures still come from the developer and partner evaluations; there is no public independent reproduction covering the complete cybersecurity capability set.
Direct sources: Mistral, “Mistral Large 4 (Le Chonk) is now open weight”, 29 October 2026, https://mistral.ai/news/mistral-large-4-oss/; Hugging Face, “Welcome Le Chonk: Mistral Large 4 on Hugging Face”, 29 October 2026, https://huggingface.co/blog/mistral-large4; Reuters, “Mistral AI publicly releases Large 4, targets cybersecurity users”, 29 October 2026, https://www.reuters.com/legal/litigation/mistral-ai-publicly-releases-large-4-targets-cybersecurity-users-2026-10-29/
Gemini 4 Carbon begins a restricted rollout to cyber defenders
Google announced Gemini 4 Carbon on 28 October and is starting access this week for a subset of cyber defenders before gradually adding more trusted partners through the Fairwind Program. Developers, businesses and consumers do not yet have broad access, and the announcement does not provide a public API model identifier or a general-availability date. Introductory pricing is listed at US$2 per million input tokens and US$10 per million output tokens, with cached input charged at five per cent of the input price. The planned post-introductory rates are US$4 and US$20.
Google reports scores of 88.2 per cent on SEC-bench Pro, 54.8 per cent on ExploitGym and 71.0 per cent on CWE-bench, as well as a zero attack success rate in its indirect prompt-injection evaluation. Artificial Analysis independently gives Carbon a score of 59 on Intelligence Index v5 and confirms that it ranks first on that composite index, although it does not lead every coding and knowledge-work component. The evaluator also confirms that Carbon is rolling out only to selected users and is not publicly available. Google says it is participating in the US Government’s voluntary pre-release testing process and is continuing to collect feedback and adjust guardrails before wider access. No weights are public and the parameter count has not been disclosed.
Direct sources: Google, “Gemini 4 Carbon: our next frontier model”, 28 October 2026, https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/; Artificial Analysis, “Gemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index v5”, 28 October 2026, https://artificialanalysis.ai/articles/gemini-4-carbon-new-number-1
GitHub Agentic Workflows reaches general availability
GitHub announced general availability for the open-source GitHub Agentic Workflows project on 28 October, ending the public preview that began in June. Developers can describe tasks such as issue triage, CI failure analysis, documentation updates and test writing in natural-language Markdown files, which the tool compiles into standard GitHub Actions YAML. Workflows reuse existing runner groups and policy constraints and can run with GitHub Copilot or another coding agent. The tool is now available for GitHub Free, Team and Enterprise, although the default Copilot CLI engine still requires a Copilot licence.
Changes since preview include Actions Runner Controller and Docker-in-Docker support, a budget forecasting command that is no longer experimental, model routing, additional workflow templates and stronger isolation options. Agents use read-only permissions by default and run in a sandboxed container behind the Agent Workflow Firewall. Outputs pass through safe-output validation, and a dedicated threat-detection job scans proposed changes. Users can choose Cloud Hypervisor or NVX micro-VM isolation. GitHub retains the requirement for human review: general availability does not remove the need to review agent-generated changes and validate their correctness.
Direct sources: GitHub, “GitHub Agentic Workflows generally available”, 28 October 2026, https://github.blog/changelog/2026-10-28-github-agentic-workflows-generally-available/; GitHub, “github/gh-aw releases”, as at 29 October 2026, https://github.com/github/gh-aw/releases
Sources
Mistral, “Mistral Large 4 (Le Chonk) is now open weight”, 29 October 2026, https://mistral.ai/news/mistral-large-4-oss/
Hugging Face, “Welcome Le Chonk: Mistral Large 4 on Hugging Face”, 29 October 2026, https://huggingface.co/blog/mistral-large4
Reuters, “Mistral AI publicly releases Large 4, targets cybersecurity users”, 29 October 2026, https://www.reuters.com/legal/litigation/mistral-ai-publicly-releases-large-4-targets-cybersecurity-users-2026-10-29/
Google, “Gemini 4 Carbon: our next frontier model”, 28 October 2026, https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/
Artificial Analysis, “Gemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index v5”, 28 October 2026, https://artificialanalysis.ai/articles/gemini-4-carbon-new-number-1
GitHub, “GitHub Agentic Workflows generally available”, 28 October 2026, https://github.blog/changelog/2026-10-28-github-agentic-workflows-generally-available/
GitHub, “github/gh-aw releases”, as at 29 October 2026, https://github.com/github/gh-aw/releases
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.