Previewing GPT-5.6 Sol: a next-generation model
OpenAI Blog · 2026-06-26
OpenAI announces a limited, government-coordinated preview of GPT-5.6 Sol, its most capable model featuring new max-reasoning and ultra multi-agent modes, citing advanced cybersecurity capabilities that approach but do not cross the Cyber Critical threshold under its Preparedness Framework.
Appears in
Extraction
Topics: gpt-5-6frontier-model-releasesai-safetycybersecurity-aiopenai
Claims
- GPT-5.6 Sol sets a new state of the art on Terminal-Bench 2.1 for long-horizon command-line coding workflows.
- GPT-5.6 Sol achieves performance competitive with Mythos Preview on ExploitBench using approximately one-third of the output tokens.
- OpenAI dedicated over 700,000 A100-equivalent GPU hours to automated red teaming aimed at finding universal jailbreaks.
- GPT-5.6 Sol identified exploitation primitives in Chromium and Firefox but did not autonomously produce a functional full-chain exploit under tested conditions.
- OpenAI considers the government-coordinated phased release a short-term step and explicitly opposes it becoming the long-term default for model releases.
Key quotes
We don't believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.
GPT-5.6 Sol is better at helping people find and fix vulnerabilities than reliably carrying out end-to-end attacks.
We dedicated over 700,000 A100-equivalent GPU hours to automated red teaming aimed at finding universal jailbreaks: attacks that can work across many prompts or contexts, not just one narrow setting.