OpenAI Launches GPT-5.6, Sol Leads Charge

OpenAI officially released its GPT-5.6 model family for general availability on July 9, 2026. This launch, following a limited preview, introduces three models: Sol, the new flagship; Terra, a balanced option for daily use; and Luna, the most cost-efficient variant.

GPT-5.6 Sol establishes a new standard for AI intelligence and efficiency. It achieves state-of-the-art results in coding, knowledge work, cybersecurity, and scientific applications, outperforming previous and competing frontier models with fewer tokens and lower estimated costs. This translates to superior performance per dollar.

Efficiency and Performance Redefined

The models deliver more intelligence per token and stronger performance per dollar. The 'ultra' setting, for instance, coordinates multiple agents across parallel workstreams, accelerating complex tasks.

On the Agents’ Last Exam, an evaluation spanning 55 professional workflows, GPT-5.6 Sol achieved a score of 53.6, surpassing Claude Fable 5 by 13.1 points. Even at medium reasoning, it beat Fable 5 by 11.4 points at roughly one-quarter the estimated cost. This large language model efficiency extends to Terra and Luna, which outperform Fable 5 at about one-sixteenth the cost.

GPT-5.6 Sol also excels in coding, setting a new state-of-the-art score of 80 on the Artificial Analysis Coding Agent Index. This is 2.8 points above Fable 5, accomplished with less than half the output tokens, under half the time, and at approximately one-third less cost. OpenAI GPT-5.6 Sol performance is further bolstered by new features like Programmatic Tool Calling in the Responses API, streamlining tool-heavy tasks with fewer tokens and model round trips.

Beyond efficiency, GPT-5.6 demonstrates a significant leap in design judgment. It can create tasteful, functional interfaces from high-level directions and refine rendered results, not just generate code.

End-to-End Knowledge Work and Safeguards

For professional tasks, GPT-5.6 processes messy context from various documents and platforms like Slack and Microsoft 365, converting it into expert-level artifacts. GPT-5.6 Sol achieved new state-of-the-art results on BrowseComp at 92.2% and OSWorld 2.0 at 62.6%.

The model family also enhances presentations, documents, and spreadsheets, producing more polished and accurate outputs. It can infer and apply design systems consistently, crucial for repeatable knowledge work.

OpenAI emphasizes robust safeguards with GPT-5.6, designed to be resilient against misuse without limiting legitimate applications. The models underwent extensive evaluation, combining human red teaming with large-scale automated testing, prior to general availability.