SignalFrontier releaseSG-0021
OpenAI released GPT-6.1 Sol, a cheaper model it also rates Critical for cyber capability
OpenAI says GPT-6.1 Sol delivers capabilities comparable to GPT-6 Astra with more speed and lower cost. OpenAI treats it as Critical in cybersecurity and High in biology and chemistry under its Preparedness Framework, applies Astra's safeguards, and is phasing advanced cyber access through Daybreak. On ExploitGym it built working exploits in 35.1% of attempts, against 42.4% for Astra. In alignment tests it never tried to bypass an automated reviewer, but it pushed past low-stakes warnings in 23.5% of rollouts, compared with 17.4% for Astra.
Why it matters
A second model rated Critical for cyber capability, offered more cheaply and cleared for public launch, widens access to exploit-building skill. That puts more weight on safeguards beyond model refusals, and OpenAI reports modest regressions in agentic cyber-safety tests. Its higher rate of pushing past warnings was measured in tests without system-level controls; it is not an observed incident.
Sources
Read the reporting
1 source. Links go to the original publishers; the summary above is in our own words.
- primaryAddendum to GPT-6 Astra System Card: GPT-6.1 SolOpenAI · Sept. 29, 2026deploymentsafety.openai.com/gpt-6-1-sol/
Signals
More from the news desk
Cite and share
Use this signal
Citation
Paperclip Index. “OpenAI released GPT-6.1 Sol, a cheaper model it also rates Critical for cyber capability.” Signal SG-0021. Reported Sept. 29, 2026; updated Oct. 7, 2026. https://paperclipindex.com/signal/SG-0021