GPT-5.6 Sol
OpenAI's GPT-5.6 Sol is a new flagship model offering enhanced reasoning, state-of-the-art coding and cybersecurity capabilities, and significant improvements in biology workflows, currently in limited preview.
The full picture
GPT-5.6 Sol is the flagship model in OpenAI’s new GPT-5.6 series, which also includes Terra and Luna. Announced on June 26, 2026, it introduces a novel “max reasoning effort” feature designed to allow the model to reason more deeply and an “ultra mode” that uses subagents for accelerating complex tasks. The model sets a new state of the art on Terminal-Bench 2.1, a benchmark for command-line coding workflows, indicating strong capabilities in planning, iteration, and tool coordination for software development. Furthermore, GPT-5.6 Sol demonstrates significantly stronger cyber capabilities, excelling in areas like vulnerability research and exploitation, and shows broad improvements in biology workflows. OpenAI emphasizes a robust safety stack built into GPT-5.6 Sol, with strengthened protections against high-risk activities and misuse, aiming to balance legitimate defensive work with constraints on prohibited offensive use. While currently in a limited preview for trusted partners, OpenAI plans for general availability in the coming weeks. Pricing information provided includes details on cache writes and reads, indicating a focus on optimizing cost for repeat operations. Terra and Luna variants are also part of this series, with Terra offering competitive performance at lower cost and Luna providing strong capability at the lowest cost point. This release signifies a major step forward for OpenAI in delivering highly capable and versatile AI models for a range of demanding applications.
What it can do and where it shines
Key capabilities
- Advanced Reasoning
- Coding
- Cybersecurity
- Biology Workflows
Best use cases
- Complex software engineering
- Cybersecurity vulnerability research and exploitation
- Advanced reasoning tasks
- Biology research
Pricing
Cache writes billed at 1.25x uncached input rate, cache reads receive 90% discount.
What we liked
- New max reasoning effort for deeper thought
- Ultra mode leveraging subagents for complex work
- State-of-the-art in command-line coding workflows (Terminal-Bench 2.1)
- Significant improvements in cyber capabilities for defensive work
- Robust safety stack
- Broad improvements in biology workflows
Watch out for
- Currently in limited preview, not broadly available yet
- Pricing details for non-cached operations are opaque
Limitations
Limited preview access
Full details on multimodal capabilities not explicitly stated
Comprehensive pricing information not fully public
Editorial take
GPT-5.6 Sol represents a significant leap in OpenAI’s model capabilities, particularly for complex reasoning, coding, and cybersecurity, positioning it as a top-tier choice for enterprises requiring advanced autonomous agentic features.