AI Risk Wire

Models

For each AI model version: the capability levels its provider states it reached, the outside testers who tested it, and the incidents that name it.

Also: compare models side by side · statements that a capability level was crossed · providers · practices providers document · safety news.

18 of 78 model versions

ModelLevels the provider states it reachedOutside testersIncidents naming it
GPT-6.1 Sol
OpenAI · 29 September 2026
None stated None named 1 incident
AI acting on its own
GPT-Image-2.5-Flare
OpenAI · 8 September 2026
None stated None named None
GPT-6 Astra
OpenAI · 3 September 2026
High; Critical Epoch AI, Irregular, Redwood Research, SecureBio and 1 more, Academic None
GPT-6 Sol
OpenAI · 3 September 2026
None stated None named None
GPT-5.6 Sol
OpenAI · 9 July 2026
None stated Epoch AI, FAR.AI, Irregular, METR and 3 more, Academic 3 incidents
AI acting on its own
GPT-5.6 Luna
OpenAI · 9 July 2026
None stated Academic None
GPT-5.6 Terra
OpenAI · 9 July 2026
None stated None named None
GPT-Rosalind-5.5
OpenAI · 3 June 2026
High None named None
GPT-5.5 Instant
OpenAI
High None named None
GPT-5.5
OpenAI · 23 April 2026
High Apollo Research, CAISI (NIST), Epoch AI, Irregular and 7 more None
GPT-5.5 Pro
OpenAI · 23 April 2026
None stated SecureBio None
GPT-5.4 Thinking
OpenAI · 5 March 2026
High Irregular None
GPT-5.4 mini
OpenAI · 5 March 2026
None stated Apollo Research, SaferAI, Transluce, Academic None
GPT-5.3-Codex
OpenAI · 5 February 2026
High Apollo Research, CAISI (NIST), Irregular, UK AI Security Institute None
GPT-5
OpenAI
High Apollo Research, CAISI (NIST), Epoch AI, FAR.AI and 6 more, Academic None
o3
OpenAI · 16 April 2025
None stated METR, RAND (CAST / Project Canary), Redwood Research, SecureBio None
o4-mini
OpenAI · 16 April 2025
None stated None named None
Operator
OpenAI · 23 January 2025
None stated None named None