OpenAI Launches GPT-5.6 Preview With Three Model Tiers, Outperforming Anthropic's Latest Across Key Benchmarks
Summary
OpenAI launches a limited preview of GPT-5.6 with three model tiers — Sol, Terra, and Luna — outperforming Anthropic's latest models across key benchmarks while debuting its strongest safety and cybersecurity features yet, as general availability approaches in the coming weeks.
Key Points
- OpenAI debuts a limited preview of GPT-5.6, featuring three model tiers — Sol, Terra, and Luna — targeting different levels of task complexity and price points, with general availability expected in the coming weeks.
- On TerminalBench 2.1, GPT-5.6's Sol and Ultra modes outperform Claude Mythos 5, Terra surpasses Claude Fable 5, and Luna edges past Opus 4.8, positioning OpenAI ahead of Anthropic at a time when US government export controls are restricting Anthropic's model releases.
- All three GPT-5.6 models come equipped with OpenAI's most robust safety safeguards to date, with a strong emphasis on cybersecurity capabilities including code review, defensive testing, and patch development, while questions linger about whether incremental AI upgrades will meaningfully impact everyday users.