GPT-5.6: Frontier intelligence that scales with your ambition

Source: OpenAI Date: 2026-07-09 URL: https://openai.com/index/gpt-5-6

Summary

GPT-5.6 GA’d globally across ChatGPT, Codex, and the API on 2026-07-09, closing the ~13-day preview window opened 2026-06-26. Three tiers ship with public pricing: Sol ($5/$30 per 1M, flagship, hardest coding/security problems), Terra ($2.50/$15, high-volume business tasks), and Luna ($1/$6, fast everyday work). OpenAI’s own positioning claims Terra performs just above Fable 5 and Luna outperforms Opus 4.8, both at roughly a third of the time, half the output tokens, and a quarter the cost.

Implications

• Closed-frontier clock. This is the clock firing on schedule — GA confirms rather than surprises, since Codex v0.143.0 had already integrated Sol/Terra/Luna via Bedrock support the day before GA. The tiering structure (flagship/business/fast) plus direct cost-and-latency comparisons against Anthropic’s own lineup is a structured competitive response to the open-weight wave, not just a benchmark bump.
• Agent-layer convergence. OpenAI naming its own tiers against competitor models by name in launch copy is a new level of explicit positioning — worth watching whether Anthropic or others respond in kind rather than with benchmark tables alone.
• AI adoption in enterprise. Terra’s “high-volume business tasks like customer support, internal tools and document analysis” framing is a direct pitch at the enterprise mid-tier, the same segment competitors are courting on cost-per-task rather than raw capability.