OpenAI's next-generation family — Sol, Terra, and Luna. Sol is OpenAI's strongest model to date, with a new max reasoning effort and an ultra sub-agent mode. Also searched as ChatGPT 5.6.
Update — July 1, 2026: OpenAI has published official API pricing for all three models and Terminal-Bench 2.1 scores — verified below. Still unpublished: context windows, SWE-Bench, and the wider benchmark suite. Access remains a limited preview (~20 partner orgs); general release across ChatGPT, Codex, and the API is promised in the coming weeks. No invented specs on this page — every number is confirmed.
GPT-5.6 ships as a family, not a single model — pick the tier that fits the job.
OpenAI's strongest model to date. Agentic gains in coding, biology, and cybersecurity, a new "max" reasoning effort, and an "ultra" mode that spins up sub-agents for complex work.
The everyday workhorse: competitive with GPT-5.5 quality at exactly half the cost. The likely default for most production workloads once it's generally available.
OpenAI's fastest, lowest-cost model in the 5.6 family — built for latency-sensitive and high-throughput tasks where speed and price matter more than peak reasoning.
The confirmed upgrades over GPT-5.5, straight from OpenAI's announcement.
A new effort level above the existing tiers, giving Sol more room to think on the hardest problems — for agentic coding, scientific reasoning, and long-horizon planning.
For complex work, Sol can run as a coordinator that delegates to sub-agents — closing the gap with multi-agent setups (a direction Anthropic shipped natively in Opus 4.8).
Sol sets a new state of the art on Terminal-Bench 2.1, which tests real command-line workflows that require planning, iteration, and tool coordination.
On GeneBench v1, Sol beats GPT-5.5 while using fewer tokens — a signal of more efficient reasoning, not just a bigger model.
GPT-5.6 ships with OpenAI's most robust safety stack to date, with strengthened protections around higher-risk activity, sensitive cyber requests, and repeated misuse.
Sol, Terra, and Luna span flagship, balanced, and budget — so you can match the model to the job instead of paying flagship rates for everything.
GPT-5.6's full specs aren't public yet, so here's the confirmed positioning against the models you can use right now.
| Dimension | GPT-5.6 Sol | Today's frontier (GPT-5.5 · Opus 4.8 · Gemini 3.1 Pro) |
|---|---|---|
| Status | Limited preview (~20 partner orgs) | Generally available |
| Agentic coding (Terminal-Bench 2.1) | 88.8% — 91.9% in Sol Ultra mode (new SOTA) | GPT-5.5: 88.0% · Claude Mythos 5: 88.0% |
| Reasoning modes | New "max" effort + "ultra" sub-agents | Opus 4.8: native parallel subagents |
| Pricing (per 1M tokens) | $5 in / $30 out — published Jul 1, 2026 | GPT-5.5: $5/$30 · Opus 4.8: $5/$25 |
| Context window | Not published | GPT-5.5: 1M · Opus 4.8: 1M · Gemini 3.1 Pro: 2M |
Rival figures are verified UseRightAI catalog data. GPT-5.6 numbers are intentionally left blank until OpenAI publishes them.
Where it stands at announcement — and what OpenAI says comes next.
Building today? Until GPT-5.6 is generally available with published pricing, ship on a model you can rely on now — see the best AI for coding and the live API pricing table. We'll update this page with GPT-5.6 access details the moment they go live.
GPT-5.6 is OpenAI's next-generation model family, announced June 26, 2026. It ships as three models: Sol (the flagship and OpenAI's strongest model to date), Terra (a balanced model competitive with GPT-5.5 at roughly half the cost), and Luna (the fastest, lowest-cost option). It launched as a limited preview to select partners, with broader ChatGPT, Codex, and API access planned in the coming weeks.
They're the three tiers of the GPT-5.6 family. Sol is the flagship — strongest at agentic coding, biology, and cybersecurity, with a new 'max' reasoning effort and an 'ultra' sub-agent mode. Terra is the balanced everyday model, matching GPT-5.5 quality at about 2x lower cost. Luna is the fast, cheap model for latency-sensitive and high-volume work.
OpenAI announced GPT-5.6 on June 26, 2026. At launch it is a limited preview available only to a small group of trusted partners; OpenAI says general availability across ChatGPT, Codex, and the API will follow in the coming weeks.
OpenAI published official API pricing on July 1, 2026. Per 1M tokens: GPT-5.6 Sol costs $5 input / $30 output (same as GPT-5.5), Terra costs $2.50 input / $15 output (exactly half of GPT-5.5), and Luna costs $1 input / $6 output. Prompt-cache reads are discounted 90%. Note the models are still in limited preview — you can't pay these rates yet unless you're a preview partner.
Not for everyone yet. GPT-5.6 is in a limited preview restricted to a small group of trusted partners. OpenAI plans to roll it out more broadly through ChatGPT, Codex, and the API in the coming weeks. There is no public API model ID or ChatGPT toggle for general users at announcement.
GPT-5.6 Sol beats GPT-5.5 on Terminal-Bench 2.1 (88.8% vs 88.0%, and 91.9% in Sol Ultra mode) at the same $5/$30 price, improves on GeneBench v1 while using fewer tokens, and adds 'max' effort plus an 'ultra' sub-agent mode. Terra is the value story: near-5.5-class agentic coding (82.5%) at exactly half the price. But GPT-5.5 remains the model you can actually use at scale today — GPT-5.6 is still preview-only, with context windows and full benchmarks unpublished.
'Ultra' is a new mode in which the model coordinates sub-agents to tackle complex, multi-step work — delegating pieces of a task and combining the results. It pairs with a new 'max' reasoning effort for the hardest problems. It's conceptually similar to the native parallel-subagent approach Anthropic shipped in Claude Opus 4.8.
On price they're nearly identical: GPT-5.6 Sol is $5/$30 per million tokens vs Opus 4.8's $5/$25. On benchmarks, Sol holds the new Terminal-Bench 2.1 record (88.8%, 91.9% Ultra) while Opus 4.8 holds 69.2% on SWE-Bench Pro — different tests, so no clean head-to-head yet since OpenAI hasn't released GPT-5.6 SWE-Bench or context-window numbers. And Opus 4.8 you can use today; Sol is preview-only. We'll publish a full comparison at general availability.