This observation covers technical changes formally published or newly verified from approximately 7:42 am on 5 November to 7:42 am on 6 November 2026. It includes three items.
GPT-6 Cyber becomes available to a limited group of approved customers
OpenAI released GPT-6 Cyber and its system card on 5 November. The model is a specialised GPT-6 Astra variant for approved cybersecurity work, post-trained on tasks including vulnerability discovery, exploit development and web penetration testing. It is initially available through the API and Codex to a select group of approved Daybreak Red customers; expansion to every approved customer remains planned. Organisations must also satisfy identity verification, single sign-on, multi-factor authentication and API IP-allowlisting requirements. The API catalogue specifies a 1.05-million-token context window, a maximum output of 128,000 tokens, and text and image input. No public weights are available.
Under its own Preparedness Framework, OpenAI classifies the model as Critical for cybersecurity, High for biological and chemical capabilities, and below High for AI self-improvement. OpenAI reports that its researchers used the model to identify 81 previously unknown Firefox vulnerabilities classified by Mozilla as high severity and to develop working exploits for 948 Linux kernel vulnerabilities. Reuters confirmed that OpenAI reported these figures and the limited access state, but the counts remain OpenAI results. The system card includes external evaluation by Irregular, while there is no complete public independent reproduction of the model's capabilities and safety boundaries.
METR publishes its assessment of AI R&D acceleration at Anthropic
On 5 November, METR published results from its high-access assessment at Anthropic between 12 August and 29 October. This is the first substantive output following the embedded-evaluator access arrangement disclosed in October. METR estimates that Anthropic's model capabilities on well-specified tasks have improved at about 1.9 times the earlier rate since October 2025, with a 95 per cent confidence interval of 1.5 to 2.8 times. It judges that most of the acceleration is likely attributable to AI being used in R&D rather than faster growth in training compute. The report also says AI had not yet made highly significant, fully autonomous contributions to the most important stages of model training.
The findings are METR's synthesis of internal benchmarks, interviews and data from unreleased models, rather than a reproducible experiment on a public dataset. Anthropic redacted most of the evidence METR considered most informative. METR also cautions that the result may apply mainly to tasks that are easy to grade automatically and that historical comparisons may overstate or understate the change. Reuters independently reported the assessment period, the 1.9-times estimate and these limitations.
Third-party testing follows the GPT Image 2.6 release
OpenAI released GPT Image 2.6 Sunburst and Flare on 4 November through ChatGPT, Codex and the Image API. Compared with the 2.5 family, the release adds complexity control, markup-guided editing, SVG output and animated GIF generation. Sunburst is intended for detail and editing consistency, while Flare is optimised for faster generation. OpenAI reports that generation is 50 per cent faster overall and that Flare can take about six seconds at low quality; these remain developer figures. The system card retains the core 2.5 safety stack and adds safeguards for GIF creation.
On 5 November, OmniaKey published hands-on API tests that reproduced a looping GIF, text posters and a product-photo edit. It also recorded that Batch, the Responses API and Chat Completions do not support the two models, and identified limits such as an undisclosed total pixel budget for GIF requests. This is testing by a model-service operator rather than a standardised independent benchmark. Its small sample does not verify OpenAI's overall 50 per cent speed claim.
Sources
OpenAI, GPT-6 Cyber System Card, 5 November 2026, https://deploymentsafety.openai.com/gpt-6-cyber
OpenAI, GPT-6-Cyber Model, 5 November 2026, https://developers.openai.com/api/docs/models/gpt-6-cyber
Reuters, OpenAI launches GPT-6 Cyber with 'Critical' cybersecurity rating, 5 November 2026, https://www.reuters.com/technology/openai-launches-gpt-6-cyber-with-critical-cybersecurity-rating-2026-11-05/
METR, Independent Assessment of AI R&D Acceleration at Anthropic in October 2026, 5 November 2026, https://metr.org/blog/2026-11-05-anthropic-ai-rd-acceleration-assessment/
Reuters, Anthropic's models improving nearly twice as fast with help from AI, study finds, 5 November 2026, https://www.reuters.com/technology/anthropics-models-improving-nearly-twice-fast-with-help-ai-study-finds-2026-11-05/
OpenAI, Introducing Images 2.6, 4 November 2026, https://openai.com/index/introducing-images-2-6/
OpenAI, ChatGPT Images 2.6 System Card, 4 November 2026, https://deploymentsafety.openai.com/chatgpt-images-2-6
OmniaKey, GPT Image 2.6 Review: GIF, SVG, Cost and Who It's For, 5 November 2026, https://omniakey.com/blog/gpt-image-2-6-review
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.