OpenAI launches GPT-5.6 family with three models and an ultra mode

On July 9, 2026, OpenAI made the GPT-5.6 family generally available. The lineup includes the flagship GPT-5.6 Sol, the balanced GPT-5.6 Terra, and the most affordable model, GPT-5.6 Luna. All three arrived in ChatGPT, Codex, and the OpenAI API, with a gradual global rollout over the following 24 hours.

Which models are in the lineup?

The family spans three performance tiers. Sol targets the hardest tasks: coding, professional analysis, scientific research, cybersecurity, and long-running agentic workflows. Terra balances capability and cost, while Luna is built for fast, high-volume workloads.

• GPT-5.6 Sol costs $5 per million input tokens and $30 per million output tokens
• GPT-5.6 Terra costs $2.50 per million input tokens and $15 per million output tokens
• GPT-5.6 Luna costs $1 per million input tokens and $6 per million output tokens
• Ultra mode coordinates four agents in parallel by default
• The API offers Programmatic Tool Calling and a beta of the multi-agent mode

The core idea behind the update is to get more useful work out of every token. OpenAI claims GPT-5.6 Sol matches or beats previous and competing models while using fewer tokens and less time. Terra and Luna are meant to bring a similar level of automation to everyday tasks at a lower price.

What changed for developers?

GPT-5.6 Sol is now OpenAI's strongest coding model. It scored 80 points on the Artificial Analysis Coding Agent Index and 88.8% on Terminal-Bench 2.1. Ultra mode pushed the Terminal-Bench result to 91.9% by coordinating several parallel workstreams.

The model also gains new tool-use capabilities. In the Responses API, Programmatic Tool Calling lets GPT-5.6 write and run small in-memory programs, process intermediate data, and pass only the relevant results back to the model. That cuts the number of model calls and the amount of context in tool-heavy tasks.

Max mode gives the model more time to reason, explore alternatives, and fix its mistakes. Ultra goes further: several agents work on parts of a complex task in parallel, and GPT-5.6 then merges their results. For developers, this means they can build more autonomous systems without scripting every step by hand.

Where does the model improve?

Beyond code, OpenAI highlights work with documents, presentations, and spreadsheets. GPT-5.6 follows templates, slide structure, typography, spacing, and formatting rules more precisely. The model inspects the rendered output and can fix visual and functional flaws, rather than just generating source code.

On BrowseComp, GPT-5.6 Sol at the maximum reasoning level scored 92.2%, and on OSWorld 2.0 it scored 62.6%. On ExploitBench 2, the model hit 73.5%, compared with 47.9% for GPT-5.5 at a comparable output budget. In science tasks, OpenAI reports higher scores in biology, the natural sciences, and chemistry.

On cybersecurity, the company tightened restrictions at the same time. GPT-5.6 supports defensive work such as vulnerability analysis, code remediation, threat modeling, and blue teaming. Only vetted users and organizations get access to the most sensitive capabilities, through the Trusted Access for Cyber program.

What does it mean?

GPT-5.6 isn't just an upgrade in answer quality; it's OpenAI's attempt to turn the model into a more independent executor of complex work. The three pricing tiers, programmatic tool calls, and parallel agents should lower the cost of long-running workflows, though the top modes will require more tokens and compute.