Use OpenAI's flagship GPT-5.6 model for complex, multi-step professional work, from agentic coding to large-document analysis.
GPT-5.6 Sol is OpenAI's highest-capability model in the GPT-5.6 family, designed for complex, multi-step professional work such as agentic coding, security review, and large-document analysis.
It sits above Terra, the balanced model, and Luna, the fast and affordable model, in the GPT-5.6 tiering. Choose Sol when task reliability, long-horizon reasoning, and deep workflow execution matter more than minimum latency or cost.
OpenAI adopted an astronomical naming convention for the GPT-5.6 family to clearly distinguish each model's relative size, compute power, and cost efficiency.
Sol, meaning the Sun, is the largest, most capable, and most computationally intensive model. It is built for complex reasoning, long-horizon problem-solving, and advanced multi-agent tasks.
Terra, meaning the Earth, is the mid-tier option designed to balance high capability with lower costs for everyday knowledge work. Luna, meaning the Moon, is the smallest, fastest, and most affordable model in the family.
GPT-5.6 Sol represents a shift in extended, multi-step problem solving. Rather than simply executing the first available solution, Sol is designed to orient itself within unfamiliar environments. Max and Ultra modes allow the model to spend significantly more time thinking before responding.
Sol operates more like a persistent, high-level software engineer than a simple code-completion tool, especially in environments like Codex and ChatGPT Work. It can trace bugs through massive, unfamiliar production codebases and carry large projects end-to-end.
Sol's advanced reasoning translates into stronger performance in rigorous STEM fields. With a 1-million-token context window, it can ingest and synthesize scientific literature, biological datasets, and complex mathematical proofs in a single prompt.
The model is strong at vulnerability research, patch development, and defensive testing. It is also explicitly restricted from completing autonomous, end-to-end offensive cyberattacks against hardened targets.
Beyond technical execution, GPT-5.6 Sol is optimized for enterprise productivity and polished knowledge work. It can show stronger design judgment when creating slide decks, reports, and structured business documents.
GPT-5.6 Sol is more autonomous: it can coordinate sub-agents and run longer, more complex plans without constant human prompting.
It is stronger on hard reasoning, with noticeable improvements on tough coding, math, and multi-step scientific problems.
It is safer but more controlled, with stricter usage policies and monitoring, especially for high-risk domains such as cybersecurity, biological research, and chemical risk analysis.
| Model | Release date | Input / output price (per 1M tokens) | Context window | Max output tokens | Tool use & agents |
|---|---|---|---|---|---|
| GPT-5.6 Sol | July 9, 2026 | $5 / $30 | 1M | 128K | Full agent stack for tool calling, code execution, browser, and computer-use workflows. |
| GPT-5.6 Terra | July 9, 2026 | $2.50 / $15 | 1M | 128K | Balanced GPT-5.6 tier with high capability at lower cost for everyday professional work. |
| GPT-5.6 Luna | July 9, 2026 | $1 / $6 | 1M | 128K | Fast and affordable GPT-5.6 tier for lightweight, repetitive, or latency-sensitive requests. |
| GPT-5.5 | April 23, 2026 | $5 / $30 | 1M | 128K | Earlier flagship with weaker long-horizon planning and complex multi-step tool use compared with GPT-5.6 Sol. |
| Claude Opus 4.7 | May 28, 2026 | $5 / $25 | 1M | 128K | Available Claude model in ChatMax AI for long-form reasoning, writing, document work, and coding assistance. |
GPT-5.6 Sol is optimized for high-value knowledge work requiring deep, persistent analysis. It is ideal for end-to-end software development in environments like Codex, complex scientific discovery, enterprise decision support, and advanced security workflows such as vulnerability detection and long-horizon cybersecurity investigations.
The Max setting grants Sol extended compute time to explore alternatives and refine complex answers. Ultra mode goes further by coordinating four AI sub-agents in parallel by default, letting them execute different parts of massive, multi-step workflows at the same time.
GPT-5.6 Sol's foundational training data has a strict knowledge cutoff of February 2026. For events or developments after that date, Sol relies on browsing and external tools to retrieve current information when those tools are available.
Choose Sol for highly complex coding, deep reasoning, and autonomous multi-agent workflows. Choose Terra as a balanced, cost-effective default for everyday professional tasks. Choose Luna for lightweight, repetitive requests where maximum speed and lowest cost are the top priorities.
OpenAI treats Sol, Terra, and Luna as high capability in cybersecurity and biological or chemical risk under its Preparedness Framework, but not at the Critical threshold. Safeguards are designed to make prohibited offensive activity more difficult, uncertain, and detectable. Human review remains essential for production risk decisions; Sol should not be the sole authority on security posture.
