• Category: AI
  • Status: confirmed
  • Sources: OpenAI announcement, OpenAI preview, HN discussion
  • Summary: OpenAI made the GPT-5.6 family generally available on 2026-07-09 across ChatGPT, the API, and Codex, after the government-coordinated testing period that the US Department of Commerce Center for AI Standards and Innovation completed before the staggered-release restriction was lifted. The family has three tiers: Sol, the flagship for complex reasoning and agentic coding, Terra, described as competitive with GPT-5.5 at half the cost, and Luna, the cheapest. GPT-5.6 adds a new maximum reasoning effort and an ultra mode that spawns subagents to parallelize complex work. Reported pricing per million tokens is Sol $5 in and $30 out, Terra $2.50 and $15, and Luna $1 and $6. OpenAI states Sol runs on Cerebras at up to 750 tokens per second in July.
  • Comments: HN commenters were skeptical of the launch benchmarks, noting the "Agents' Last Exam" chart truncates its y-axis at 30 percent to overstate the gains and compares against Fable 5 in adaptive rather than max reasoning mode. Several read the release as an incremental version bump.
  • Why it matters: OpenAI's flagship family reaching general availability on its most-used surfaces, with Terra undercutting GPT-5.5 and Sol targeting agentic coding, lands directly against this week's Grok 4.5 and SWE-1.7 releases and resets the frontier-model comparison for developers.
  • Follow-up: Watch independent benchmark reproduction, the final per-tier pricing, and the subagent ultra-mode behavior in Codex.

Send feedback on this story