GPT-5.6 beats Claude Fable 5, OpenAI calls it 'big milestone'
What's the story
OpenAI has announced a major victory over its competitor, Anthropic, with the launch of its latest AI model, GPT-5.6 Sol. The new model has outperformed Anthropic's Claude Fable 5 in an independent front-end design benchmark conducted by Design Arena, the world's first crowdsourced benchmark for AI-generated design. The results were shared by OpenAI President Greg Brockman on X (formerly Twitter), who described it as a "big milestone."
Benchmark results
'Sol' tops Design Arena leaderboard
The Design Arena leaderboard, which ranks AI models on front-end development and web design tasks using an Elo rating system, has placed GPT-5.6 Sol at the top with an Elo score of 1,353. It narrowly beats GLM 5.2 (1,351) and Claude Fable 5 (1,345). The leaderboard, updated on July 13, also features Claude Opus 4.6 Thinking among the top-performing models in this category.
Model features
Sol, Terra, and Luna: The new GPT-5.6 family
OpenAI describes Sol as its most advanced model to date, built for complex reasoning, coding, and agentic workflows. The company claims it also shows improvements in scientific research tasks such as biology and cybersecurity. Along with Sol, OpenAI is also launching Terra, a balanced model for everyday workloads, and Luna, a cost-effective model optimized for speed.
Model enhancements
Improvements in biology research and cybersecurity evaluations
The new models come with advanced capabilities such as a Maximum Reasoning mode for more complex problems and an Ultra mode that employs sub-agents to complete multi-step tasks efficiently. Beyond software development, OpenAI claims GPT-5.6 outperforms its predecessor on biology research benchmarks while consuming fewer tokens. It also stays competitive in cybersecurity evaluations without breaching its internal "Cyber Critical" safety threshold.
Safety focus
Enhanced safety measures in the new models
OpenAI has also emphasized safety in the preview release of GPT-5.6, which comes with improved safeguards against cyber abuse, sensitive biological misuse, and repeated attempts to bypass its protections. The company spent over 700,000 A100-equivalent GPU hours on automated red-teaming and extensive human expert testing to find vulnerabilities before launch. Currently available through API and Codex for select partners, GPT-5.6 will be more widely available across ChatGPT in coming weeks at different price points for input/output tokens.