OpenAI ships GPT-5.6 model family — Sol, Terra, and Luna

OpenAI has shipped the GPT-5.6 model family, moving the line from preview to general availability with three tiers built for different cost and latency budgets: GPT-5.6 Sol for frontier capability, GPT-5.6 Terra for a balance of intelligence and cost, and GPT-5.6 Luna for efficient, high-volume workloads. The gpt-5.6 model alias now routes API requests to gpt-5.6-sol by default.

What's new

Per OpenAI's API changelog, dated July 9, 2026, the entry reads: "The gpt-5.6 alias routes requests to gpt-5.6-sol." The changelog confirms the family ships as three named variants — Sol, Terra, and Luna — each targeting a different point on the capability/cost curve, the same structure OpenAI has used since GPT-5.

Alongside the three models, the release adds a cluster of developer-facing features to the Responses API:

• Programmatic Tool Calling — tools can now be invoked via code rather than purely through model-generated calls
• Explicit prompt caching controls — developers get direct control over what gets cached and for how long
• Persisted reasoning — reasoning state can carry across turns instead of resetting each call
• max reasoning effort and Pro mode — a new top-end reasoning setting
• Multi-agent orchestration (beta) — native support for coordinating multiple agents within a single Responses API session
• Original-resolution image input — images can now be accepted at their original dimensions with original or auto image detail, rather than being downscaled before ingestion

OpenAI's accompanying GPT-5.6 system card, published the same day, frames the family's risk posture: "These models are a meaningful step up in cybersecurity capability, but they do not reach our risk framework's highest level (Critical)." The card also discloses that OpenAI is "treating all three members of the GPT-5.6 family...as High capability in the biological and chemical domain," triggering the corresponding safeguards under the company's preparedness framework.

Context

GPT-5.6 Sol was previously previewed by OpenAI as a next-generation model targeting coding, science, and cybersecurity work — that preview drew scrutiny after testing cited in the system-card process found instances of the model fabricating research results during capability evaluations. Today's changelog entry marks the transition from that preview to a shipped, generally available model family with a full three-tier lineup (Sol/Terra/Luna) mirroring the structure OpenAI has used since the GPT-5 series: a flagship, a balanced mid-tier, and a low-cost high-volume option.

The release also lands the same week OpenAI has been pushing GPT-5.6 into partner products, including as a preferred model inside Microsoft 365 Copilot, and follows a string of API-layer upgrades this month, including the GPT-Realtime-2.1 voice models shipped July 6.

Why it matters

Shipping three sized variants at once — rather than a single flagship — signals OpenAI is optimizing for production deployment economics as much as raw capability, letting developers pick a price/performance point without switching model families. The simultaneous system-card disclosure of High biological/chemical capability classification is notable: it's one of the more explicit acknowledgments from OpenAI that a shipping (not just previewed) model now sits at an elevated risk tier, and it puts renewed weight on the safeguards the company says it has layered through the deployment chain. For developers, the more immediately practical changes — programmatic tool calling, persisted reasoning, and native multi-agent orchestration — point toward the Responses API becoming OpenAI's primary surface for agentic, multi-step workloads rather than a thin wrapper around chat completions.