Overview
GPT-5.6 is OpenAI's July 9, 2026 model family for ChatGPT, Codex, and the OpenAI API. It introduces Sol as the flagship tier, Terra as the balanced tier, and Luna as the fastest, lowest-cost tier. Unlike the preceding GPT-5.4 mini and nano split, the family gives teams durable capability tiers while retaining a shared generation. It is aimed at people who need to choose deliberately between demanding agentic work, everyday professional tasks, and high-volume API execution.
The practical upgrade question is not simply whether Sol is stronger. It is whether a workflow needs Sol's highest-effort reasoning and ultra setting, Terra's lower-cost general capability, or Luna's lowest listed token price. OpenAI makes all three available across ChatGPT Work, Codex, and the API, but plan access and effort controls differ by surface.
What's New
Three capability tiers
Sol targets the hardest coding, knowledge-work, science, and cybersecurity tasks; Terra is OpenAI's balanced everyday tier; Luna is the fastest and most economical tier. The tiers are not separate product generations: they share GPT-5.6 but let a team assign expensive reasoning only where it adds value. ChatGPT Work and Codex users can select model tiers and effort settings according to their plan.
Reasoning and coding improvements
OpenAI positions Sol as its strongest coding model to date. Its release reports an 80 score on the Artificial Analysis Coding Agent Index at maximum reasoning, alongside stronger Terminal-Bench 2.1 and DeepSWE results than previous GPT models. These are vendor-reported release figures, not a guarantee for a particular codebase; teams should evaluate their own repository tasks before replacing a production default.
Better long-running work
OpenAI reports 53.6 on Agents' Last Exam for GPT-5.6 Sol and emphasizes tool coordination, progress monitoring, and intermediate-program execution. The operational change is important for teams building long-running agents: the model can write lightweight programs to coordinate tools rather than treating every intermediate step as a separate conversational turn.
Multi-agent and tool workflows
The Responses API adds Programmatic Tool Calling and a beta multi-agent capability, allowing developers to coordinate concurrent subagents and synthesize their work.
Availability & Access
GPT-5.6 began a global rollout on July 9, 2026 and OpenAI said availability would continue over the following 24 hours. In ChatGPT, Sol is available to Plus, Pro, Business, and Enterprise users at medium and higher effort; Sol Pro is limited to Pro and Enterprise. In ChatGPT Work and Codex, Free and Go users receive Terra, while Plus and higher plans can choose Sol, Terra, and Luna. API access is separate from a ChatGPT subscription and follows OpenAI's supported-country and account requirements.
Pricing & Plans
API pricing per million tokens is $5 input/$30 output for Sol, $2.50/$15 for Terra, and $1/$6 for Luna. Cached input receives OpenAI's stated 90% discount; GPT-5.6 also introduces a 30-minute minimum cache life and explicit cache breakpoints. ChatGPT plan pricing is separate and varies by region, so this page does not treat API token rates as a subscription price.
Migration Guide
Start by routing a representative set of production tasks to Terra and Sol rather than replacing a previous default globally. Compare success rate, latency, tool-call count, and total token spend; the strongest tier is not automatically the most economical choice for routine classification or extraction. API teams using cached prompts should test explicit cache breakpoints because GPT-5.6 changes cache-life and cache-write billing rules.
Best For
- Teams building agentic coding or knowledge-work workflows.
- Developers who need a lower-cost model tier alongside a flagship model.
- ChatGPT and Codex users who can select effort level per task.
FAQ
When was GPT-5.6 released?
OpenAI announced general availability on July 9, 2026.
Where is GPT-5.6 available?
OpenAI lists ChatGPT, Codex, and the OpenAI API, subject to plan and rollout availability.
Which GPT-5.6 model should I use?
Use Sol for the hardest coding or research tasks, Terra for balanced everyday work, and Luna when cost and latency matter most. Validate that choice on representative tasks instead of choosing from benchmark scores alone.
Does GPT-5.6 change API pricing?
Yes. OpenAI lists Sol at $5/$30, Terra at $2.50/$15, and Luna at $1/$6 per million input/output tokens. Cached input is discounted, but cache writes have a separate rate.
