OpenAI launches GPT-5.6 Sol, Terra and Luna

What OpenAI shipped and when

OpenAI released the GPT-5.6 family for general availability on 9 July 2026 across ChatGPT, Codex and the OpenAI APIverifiedVerified Source: openai.com.^[1] The rollout delivered three distinct variants in a single announcement, the most granular product segmentation OpenAI has made to date.

GPT-5.5, released on 23 April 2026, had served as the backbone of everyday ChatGPT interactions.^[1] GPT-5.6 builds on that foundation with incremental improvements in reasoning, efficiency and performance across coding, knowledge work, cybersecurity and scientific applications.

Sol, Terra and Luna said

Sol is the flagship, built for heavy reasoning, coding and science.^[1] Terra sits in the middle, a balanced everyday model priced at roughly half the cost of GPT-5.5, while Luna targets cost-sensitive, high-volume workloads where raw throughput matters more than peak intelligence.^[1]

Accessing Sol inside ChatGPT requires a paid subscription and a deliberate choice. GPT-5.5 Instant remains the default model for fast, everyday responses; paid users must select medium or higher reasoning-effort settings to reach Sol.verifiedVerified Source: help.openai.com^[2] That design keeps the standard interface snappy while giving power users the option to dial up capability.

Pricing breakdown

Sol costs $5 per million input tokens and $30 per million output tokens; Terra runs at $2.50 input and $15 output; Luna sits at $1 input and $6 output.verifiedVerified Source: openai.com^[1] The spread gives enterprises a direct lever: pay premium rates for Sol's reasoning depth, or route routine queries through Luna to protect margins.

At medium reasoning effort, Sol still outperformed Anthropic's Claude Fable 5 on Agents' Last Exam by 11.4 points, at roughly one-quarter the cost.^[1] That cost-to-performance ratio is the sharpest commercial argument OpenAI is making for the new family.

Performance versus GPT-5.5 and the competition

OpenAI said GPT-5.6 Sol scored 53.6 on Agents' Last Exam, surpassing Anthropic's Claude Fable 5 by 13.1 points.^[1] Google's Gemini series, which targets multimodal tasks, was not benchmarked in the same results.

Alex Wang, Applied AI at financial research platform Rogo, said the gains translated directly into production workloads. "GPT-5.6 is a major step forward for financial research agents. On Rogo's Big Finance Benchmark, it improved rubric quality by 6.2 points and answer accuracy by 3.6 points versus GPT-5.5. With Programmatic Tool Calling, it matched quality while using 24% fewer output tokens and completing tasks 28% faster. That combination of accuracy, speed, and efficiency is exactly what we need to scale high-quality financial analysis."^[1]

Legal software company Clio reported similar results. Angel Faus, VP of Engineering at Clio, said: "Across legal research and document workflows, GPT-5.6 is already delivering the kind of efficiency gains that change product economics. In our combined evaluation suite, it uses 14% fewer tokens while improving quality across legal research and transactional law use cases. For multi-step document analysis, Programmatic Tool Calling cuts prompt tokens by 38% with no quality loss."^[1]

What it means for businesses and developers

The tiered structure shifts how AI procurement decisions get made. Businesses previously chose between paying for a single capable model or accepting a cheaper but weaker alternative. GPT-5.6 lets teams assign workloads to the cheapest model that can handle them, routing complex analytical tasks to Sol and repetitive classification jobs to Luna.

Developers accessing the family via the OpenAI API can switch between variants by changing the model parameter, with no separate integration required.^[1] OpenAI made all three variants available through Codex simultaneously, extending the pricing choice into automated software development pipelines as of 9 July 2026.