GPT-5.6 is OpenAI's most capable model to date, and it went generally available on July 9, 2026. The headline change isn't a single model — it's three: Sol, Terra, and Luna, a flagship, a balanced mid-tier, and a fast cheap tier that share a generation but scale independently. At the top, Sol sets new highs on agentic and terminal-coding benchmarks and carries a 1.05M-token context window, which makes it the model to reach for on hard, long-running work.
Verdict: A genuinely strong frontier family. Sol leads on autonomous coding and long-horizon reasoning, and the three-tier split finally lets you match model cost to task difficulty instead of overpaying.
Best for: developers, agent builders, and teams running long, tool-heavy workflows who want to tune cost against capability.
What is GPT-5.6?
GPT-5.6 is the model family behind ChatGPT and OpenAI's API. It launched to about 20 government-approved organizations as a limited preview on June 26 before its public release on July 9, and it arrived alongside ChatGPT Work, an agent built to carry out whole jobs rather than answer one-off questions.
The design idea is that the number (5.6) marks the generation, while the names mark durable capability tiers that can advance on their own cadence. So you pick a tier for the job — top-end reasoning, balanced, or fast — and OpenAI can improve each independently over time.
Sol, Terra and Luna
Sol is the flagship, built for hard problems and long tasks it runs largely on its own. It's the one you want for complex agentic coding, deep research, and multi-step reasoning. Terra is the balanced middle: OpenAI positions it at "competitive performance to GPT-5.5 while being 2x cheaper," which makes it the sensible default for most production work. Luna is the fast, cheap tier for high-volume, lower-difficulty tasks like classification, extraction, and routing.
That spread is the real story. Instead of paying flagship rates for everything, you send simple work to Luna, everyday work to Terra, and only the genuinely hard problems to Sol. Our GPT-5.6 pricing guide works through exactly where each tier pays off.
Benchmarks and quality
Sol's benchmark results are the strongest OpenAI has posted. On Agents' Last Exam — a test of long-running professional workflows across 55 fields — Sol sets a new high of 53.6. On Terminal-Bench 2.1, which measures command-line workflows that require planning, iteration, and tool coordination, it reaches a category-leading 88.8%.
In practice, that translates to noticeably better autonomous coding and tool use. Sol plans multi-step tasks well, recovers from its own mistakes more reliably than GPT-5.5, and holds context across very long sessions thanks to the 1M-token window and 128K max output. It's not flawless — it can still over-engineer and occasionally states results with more confidence than the evidence supports — but on hard technical work it's excellent.
Agents and ChatGPT Work
GPT-5.6 leans hard into agents. New in the Responses API are programmatic tool calling — the model writes and runs small programs in memory to coordinate tools — plus multi-agent support in beta, letting one request spin up concurrent subagents and synthesize their work. There's also explicit prompt caching, persisted reasoning, a new maximum reasoning level, and a Pro reasoning mode.
On the product side, ChatGPT Work is the consumer-facing expression of that: an agent aimed at completing jobs end to end. Terra and Luna are pitched primarily at ChatGPT Work, Codex, and API users, while Sol is exposed through reasoning settings in standard ChatGPT for eligible paid plans.
Pricing and access
On the API, the three tiers are priced per million tokens: Sol at $5 input / $30 output, Terra at $2.50 / $15, and Luna at $1 / $6. Explicit prompt caching can cut input costs further on repetitive workloads.
| Tier | API price (per 1M tokens) | Best for |
|---|---|---|
| Sol | $5 in / $30 out | Hard, long-running, agentic tasks |
| Terra | $2.50 in / $15 out | Balanced everyday production work |
| Luna | $1 in / $6 out | High-volume, simple tasks |
If you use ChatGPT rather than the API, Sol shows up as a reasoning option on paid tiers — see our ChatGPT pricing guide for the Plus and Pro plans. For the full API breakdown and how caching changes the math, read the GPT-5.6 pricing guide.
Pros & cons
Pros
- Sol leads on agentic and terminal-coding benchmarks
- Three tiers let you match cost to task difficulty
- 1.05M-token context with 128K max output
- Programmatic tool calling and multi-agent (beta)
- Terra roughly halves GPT-5.5's cost at similar quality
Cons
- Sol's flagship output pricing ($30/M) is steep at scale
- Multi-agent and some features are still beta
- Full Sol access gated to eligible paid plans and API
- Can over-engineer and occasionally overstate confidence
- Tier sprawl adds a routing decision to every workload
Who it's for
Choose GPT-5.6 if you build agents, ship code with AI, or run long tool-heavy workflows and want to tune spend across tiers. Sol is the pick for the hardest problems; Terra is the smart default; Luna handles volume. Think twice if your needs are simple chat and writing — a cheaper model or a flat ChatGPT subscription may be all you need.
Deciding between the frontier labs? Read GPT-5.6 vs Claude Opus 4.8 and GPT-5.6 vs Gemini 3.5 Pro, and see where it lands in our best AI chatbots ranking.
Frequently Asked Questions
Is GPT-5.6 worth using in 2026?
Yes. It's OpenAI's most capable model family, with a 1M-token context, state-of-the-art agentic and terminal-coding benchmarks, and three tiers so you can pay for exactly the intelligence you need. Sol is the flagship; Terra and Luna cut cost sharply for lighter work.
What are Sol, Terra and Luna?
They're the three tiers of the GPT-5.6 family. Sol is the top flagship for hard, long-running tasks; Terra is balanced and matches GPT-5.5 at about half the cost; Luna is the fast, cheap tier for high-volume simple work.
How much does GPT-5.6 cost?
On the API, Sol is $5 input and $30 output per million tokens, Terra is $2.50 and $15, and Luna is $1 and $6. In ChatGPT, Sol is available through reasoning settings on eligible paid plans.
How big is GPT-5.6's context window?
Sol supports a 1.05M-token context window with up to 128K tokens of output, which is why it holds up on very long documents and multi-step agent sessions.
What is ChatGPT Work?
ChatGPT Work is an agent released alongside GPT-5.6, built to carry out whole jobs end to end rather than answer single questions. Terra and Luna are aimed primarily at ChatGPT Work, Codex, and API use.
Is GPT-5.6 better than GPT-5.5?
Yes. Sol clearly exceeds GPT-5.5 on agentic and coding benchmarks, and Terra matches GPT-5.5's quality at roughly half the price. The generation is both stronger at the top and cheaper in the middle.