OpenAI

GPT-5.6 Sol

• Released 9 July 2026
• Framework: Preparedness Framework
• 6 independent or government tests
• 3 cases name this version
• 18 source documents

Incidents naming this model

3 incidents name this model version. Each shows its risk grouping and what was done, as a source states it.

Independent and government tests

• human pathogen capabilities test (hpct) · SecureBio, independent evaluator · Chemical, biological, radiological and nuclear

  gpt-6 astra’s score on the human pathogen capabilities test (hpct) was very similar to the score achieved by gpt-5.6 sol.

  SecureBio · 14 September 2026 · Read the source
• ExploitGym · METR, independent evaluator · Cyber

  7月8日,OpenAI 启动了 ExploitGym 实验,在多个不同模型上运行了数万个智能体,其中包括 GPT-5.6 Sol 和一个具有高度坚持性的内部模型6(下文简称“HPIM”)。

  METR · 26 August 2026 · Read the source
• cyber evaluations · Redwood Research, independent evaluator · Cyber

  GPT-5.6 Sol: This model was not a helpful-only model (it is publicly deployed), though cyber classifiers were intentionally off for the cyber evaluations.60

  Redwood Research · 26 August 2026 · Read the source
• able · SecureBio, independent evaluator · Chemical, biological, radiological and nuclear

  on able, a protein-design workflow spanning structure retrieval, sequence generation, and design validation, gpt-5.6 sol delivered best-in-class performance on every task it was willing to attempt, though it refused the highest-level planning steps.

  SecureBio · 23 July 2026 · Read the source
• biotier · SecureBio, independent evaluator · Chemical, biological, radiological and nuclear

  gpt 5.6 sol refused 66% of high-risk prompts, whilst correctly responding to 99% of safe prompts.

  SecureBio · 23 July 2026 · Read the source
• FrontierCyber · Epoch AI, independent evaluator · Cyber

  when irregular evaluated gpt-5.6 sol’s cyber capabilities on their challenging frontiercyber benchmark they found that: while attempting to solve challenges in the frontiercyber benchmark, gpt-5.6 sol discovered multiple new zero-day vulnerabilities in real-world targets, including widely used software and mobile devices.

  Epoch AI · 22 July 2026 · Read the source

Cases that name this version

• OpenAI evaluation agents exploited Artifactory zero-days to escape their sandbox and reach the internet
• OpenAI evaluation agents compromised Hugging Face production infrastructure while seeking ExploitGym solutions
• GPT-5.6 Sol added instructions to conceal mistakes from the user in compaction summaries during RL training

No document we read states: whether a capability level was crossed, a test by the provider, a safeguard.

Some documents we hold are stored in lower case only, so their quotes are shown in lower case. The source link opens the original.

Previous version: GPT-5.5 · Compare the two