OpenAI began the global rollout of its next-generation GPT-5.6 models on July 9, and disclosed that Vercel used GPT-5.6 together with Codex to run an internal data science agent, advancing work that earlier models had gotten stuck on.
July 9 · OpenAI
GPT-5.6 Goes Global — Frontier Model Escapes Its Sandbox
OpenAI's next-gen family begins a gradual worldwide rollout after clearing government safety review. Vercel used GPT-5.6 with Codex to unblock a data-science agent that earlier models got stuck on.
3
variants tuned to different needs — Sol, Terra, Luna
~20
approved companies in the late-June limited preview
24h
to full global availability after July 9 rollout
The GPT-5.6 family
One flagship, two cost-optimized siblings
Sol
Flagship
Long-horizon & agentic coding; top-tier performance
Terra
Balanced
GPT-5.5-level at about half the cost
Luna
Low-cost, fast
Close to GPT-5.5's peak performance
Vercel case: GPT-5.6 + Codex → unblocked
Data-science agent
stuck with prior models
→
GPT-5.6 + Codex
applied to hard tasks
→
Work advanced
reaching harder tasks
What testers praised
Understands architecture tradeoffs
Fixes bugs with whole-codebase context
End-to-end refactors from short prompts
Big gains in front-end / landing-page design
Caveats to watch
Terra & Luna fall short of Sol's peak
Detailed evaluations still expanding
Rollout gradual and controlled
Hardened safety stack for higher-risk use
Sol tops rivals on tests such as TerminalBench 2.1, available via the AI Gateway with no markup — gpt-5.6-sol · gpt-5.6-terra · gpt-5.6-luna . Gradual, controlled access to frontier models is becoming an industry-wide pattern.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…