GPT-5.6 in ChatGPT: Sol, Terra, Luna, Access, Limits, and Pricing

GPT-5.6 is OpenAI’s new model family for complex reasoning and agentic work, but it does not replace GPT-5.5 Instant for everyday ChatGPT conversations. The family launched on July 9, 2026, with three main variants—Sol, Terra, and Luna—plus higher-compute modes for demanding tasks. This guide explains which models you can use in ChatGPT, Codex, ChatGPT Work, and the API, how much they cost, and when each option makes sense.

What is GPT-5.6?

GPT-5.6 is a three-model family available across OpenAI’s ChatGPT, Codex, ChatGPT Work, and API products. OpenAI positions Sol as its flagship reasoning model, Terra as the balanced option, and Luna as the fastest and least expensive option. In the API, the general gpt-5.6 alias routes to gpt-5.6-sol.

The launch is especially relevant for multi-step work: research, coding, document analysis, tool use, and workflows that must keep track of a long chain of instructions. OpenAI reports gains on several internal and public evaluations, but those launch figures are company-reported rather than independent testing. The best choice still depends on your own prompts, error tolerance, latency target, and budget.

GPT-5.6 Sol vs. Terra vs. Luna

GPT-5.6 Sol: the quality-first option

Sol is designed for work where the cost of a weak answer is higher than the cost of extra tokens. Good candidates include difficult code changes, cross-document synthesis, advanced planning, technical research, and tasks that require several tools. In ChatGPT, eligible users access Sol by choosing Medium, High, or—on some plans—Extra High reasoning.

GPT-5.6 Terra: the balanced production model

Terra targets teams that want much of the new family’s capability at half of Sol’s API price. It is the sensible starting point for many applications: support automation, coding assistants, data analysis, document workflows, and agents that run frequently. Terra is available in ChatGPT Work and Codex, but it is not a selectable model in the standard ChatGPT model picker.

GPT-5.6 Luna: the speed and volume model

Luna is the lowest-cost member of the family. It is a better fit for routing, classification, extraction, moderation support, short summaries, and other repeated tasks where throughput matters. It can also serve as the first stage of a workflow, passing only difficult cases to Terra or Sol.

How GPT-5.6 works in ChatGPT

Standard ChatGPT does not expose all three names as separate choices. OpenAI’s current help documentation maps the reasoning selector to different models: Instant uses GPT-5.5 Instant, while Medium and High use GPT-5.6 Sol. Pro and Enterprise users also receive additional high-compute options. This distinction matters because selecting “Instant” does not silently switch your everyday chats to GPT-5.6.

Availability and message limits can change by plan and region. OpenAI also notes that accounts may fall back to GPT-5.4 Thinking mini after a reasoning limit is reached. Check the model label before starting a long or sensitive task.

Does GPT-5.6 replace GPT-5.5 Instant?

No. GPT-5.5 Instant remains the default fast model for normal ChatGPT conversations. GPT-5.6 Sol is invoked through the reasoning choices on eligible paid plans. This creates a clearer division of labor: Instant for quick writing, questions, and everyday assistance; Sol for tasks that benefit from deliberate reasoning.

If you want the background on the default model, read our GPT-5.5 Instant guide. Users building longer workflows should also compare the new model family with ChatGPT Work, where agents, tools, and model choice matter as much as raw benchmark scores.

What changed for developers?

The GPT-5.6 API adds more control over how much computation a task receives. The family supports reasoning effort levels from none through max, depending on the model and endpoint. OpenAI recommends the Responses API for reasoning, tool use, and multi-turn workflows.

• A simple alias: gpt-5.6 routes to Sol, while explicit model names let you lock a workflow to Sol, Terra, or Luna.
• Long context: model documentation lists a 1.05-million-token context window and up to 128,000 output tokens.
• Tool-oriented features: programmatic tool calling, persisted reasoning, and a multi-agent beta are designed for longer workflows.
• Cost controls: explicit prompt caching can reduce repeated-input cost; cache reads receive a 90% discount, while cache writes are billed above standard uncached input.

A practical rollout is to evaluate the same test set on Terra and Sol first. Use Luna for repetitive stages that pass validation, then reserve Sol for exceptions or final review. This tiered pattern is usually more economical than sending every request to the flagship model.

GPT-5.6 pricing: how to control cost

At launch, API pricing ranges from $1 input and $6 output per million tokens for Luna to $5 input and $30 output for Sol. Output tokens are the larger cost driver, so concise response formats and clear stopping conditions matter. A model that finishes reliably in one attempt can also be cheaper than a smaller model that requires several retries.

1. Start with a fixed evaluation set containing easy, medium, and difficult examples.
2. Measure task success, latency, and total tokens—not model price alone.
3. Route predictable tasks to Luna, general production work to Terra, and difficult exceptions to Sol.
4. Cache stable instructions and reference material when the workflow repeats them.
5. Review failures manually before increasing reasoning effort across the entire workload.

What GPT-5.6 is best at—and where to be cautious

OpenAI presents GPT-5.6 as a step forward in reasoning, coding, tool use, and professional knowledge work. Those improvements are useful, but they do not remove the need for verification. Models can still misread ambiguous instructions, produce confident errors, or use a tool incorrectly. For financial, medical, legal, security, or publication-critical work, keep authoritative sources and human review in the loop.

The GPT-5.6 system card classifies the family as high capability in cybersecurity and biological or chemical domains and describes added safeguards and monitoring. Some higher-risk requests may therefore be blocked or reviewed. That is expected behavior, not a reliable sign that the model is malfunctioning.

Which GPT-5.6 model should you choose?

• Individual ChatGPT user: keep Instant for normal work and select Medium or High when the task requires careful reasoning.
• Developer building a new app: benchmark Terra first, then test Sol on the cases Terra misses.
• High-volume operation: validate Luna on narrow tasks with structured outputs and automatic checks.
• Research or advanced coding: use Sol with explicit sources, tests, and review criteria.
• Agent workflow: combine models by stage instead of choosing one model for the entire pipeline.

For a broader market view, see our best AI models of 2026 comparison. You can also explore the wider AI tools guide or our practical overview of GPT-Live voice workflows.

Official sources

• OpenAI: Introducing GPT-5.6
• OpenAI Help Center: GPT-5.6 in ChatGPT
• OpenAI developer guide: latest models
• OpenAI API model documentation: GPT-5.6 Terra
• OpenAI deployment safety: GPT-5.6 system card

Fact-checked against official OpenAI documentation on July 26, 2026. Plan access, limits, and pricing may change.

Frequently asked questions

Is GPT-5.6 available in ChatGPT?

Yes. Eligible paid plans can use GPT-5.6 Sol through Medium, High, and certain higher reasoning choices. Standard ChatGPT does not present Terra and Luna as separate model-picker options.

Is GPT-5.6 free?

Free and Go users do not receive GPT-5.6 Sol in standard ChatGPT. Some supported Codex access may use Terra. API usage is billed per token.

Does GPT-5.6 replace GPT-5.5 Instant?

No. GPT-5.5 Instant remains the fast default for everyday chats, while GPT-5.6 Sol handles reasoning modes on eligible plans.

Can I select GPT-5.6 Terra or Luna in normal ChatGPT?

No. OpenAI currently makes Terra and Luna available through ChatGPT Work, Codex, and the API rather than the standard ChatGPT model picker.

What is the GPT-5.6 API model name?

The gpt-5.6 alias routes to gpt-5.6-sol. Developers can also choose explicit family variants when they need a fixed balance of quality, speed, and price.

Final recommendation

GPT-5.6 is most useful as a family, not as a single automatic upgrade. Use Sol for the work that genuinely needs frontier reasoning, Terra as the default candidate for production, and Luna for validated high-volume tasks. In ChatGPT, keep GPT-5.5 Instant for speed and deliberately switch to a GPT-5.6 reasoning mode when the task justifies the extra time and limits.