GPT-5.6 beats Claude Fable 5, OpenAI calls it big milestone
OpenAI's new GPT-5.6 Sol has taken the lead over Anthropic's Claude Fable 5 in front-end design, topping the Design Arena benchmark. Alongside the benchmark win, OpenAI says the model delivers stronger coding, reasoning, safety features and improved performance across multiple AI tasks.
OpenAI is celebrating a win over rival Anthropic after Design Arena ranked its newly unveiled GPT-5.6 Sol above Claude Fable 5 in an independent front-end design benchmark. In addition to design, OpenAI says its new flagship model also surpassed Fable 5 across several of its own coding and reasoning benchmarks.
OpenAI president Greg Brockman shared the result on X, calling it a "big milestone". The post included bench mark results in Design Arena leaderboard, which evaluates AI models on front-end development and web design tasks using an Elo rating system. According to the rankings, GPT-5.6 Sol secured the No. 1 spot with an Elo score of 1353, narrowly ahead of GLM 5.2 (1351) and Claude Fable 5 (1345). The leaderboard, updated on July 13, also places Claude Opus 4.6 and Claude Opus 4.6 Thinking among the top-performing models.
What is GPT-5.6 Sol?
The benchmark results come as OpenAI introduces the GPT-5.6 family, comprising Sol, Terra and Luna, its latest AI models aimed at developers and enterprise users. The launch is currently limited to a small group of trusted partners, with a wider rollout planned in the coming weeks. OpenAI describes Sol as its most capable model yet, designed for advanced reasoning, coding and agentic workflows. The company says it also delivers improvements in scientific research tasks, including biology and cybersecurity.
Alongside Sol, OpenAI is introducing Terra, a balanced model for everyday workloads, and Luna, a lower-cost model optimised for speed. According to the company, Terra delivers performance comparable to GPT-5.5 at roughly half the cost, while Luna is its most affordable offering.
OpenAI is also introducing new capabilities with GPT-5.6. Sol includes a Maximum Reasoning mode for tackling more complex problems and an Ultra mode that uses sub-agents to complete multi-step tasks more efficiently. Beyond software development, the ChatGPT maker says GPT-5.6 performs better than GPT-5.5 on biology research benchmarks while using fewer tokens, and remains competitive in cybersecurity evaluations without crossing its internal "Cyber Critical" safety threshold.
Safety is another major focus of the preview release. OpenAI says GPT-5.6 includes stronger safeguards against cyber abuse, sensitive biological misuse and repeated attempts to bypass its protections. The company says it spent more than 700,000 A100-equivalent GPU hours on automated red-teaming, alongside extensive testing by human experts, to identify vulnerabilities before launch.
GPT-5.6 availability
GPT-5.6 is currently available through the API and Codex for select partners, with broader availability across ChatGPT planned in the coming weeks. OpenAI has priced Sol at $5 per million input tokens and $30 per million output tokens, while Terra costs $2.50 and $15, and Luna $1 and $6, respectively.
OpenAI is celebrating a win over rival Anthropic after Design Arena ranked its newly unveiled GPT-5.6 Sol above Claude Fable 5 in an independent front-end design benchmark. In addition to design, OpenAI says its new flagship model also surpassed Fable 5 across several of its own coding and reasoning benchmarks.
OpenAI president Greg Brockman shared the result on X, calling it a "big milestone". The post included bench mark results in Design Arena leaderboard, which evaluates AI models on front-end development and web design tasks using an Elo rating system. According to the rankings, GPT-5.6 Sol secured the No. 1 spot with an Elo score of 1353, narrowly ahead of GLM 5.2 (1351) and Claude Fable 5 (1345). The leaderboard, updated on July 13, also places Claude Opus 4.6 and Claude Opus 4.6 Thinking among the top-performing models.
What is GPT-5.6 Sol?
The benchmark results come as OpenAI introduces the GPT-5.6 family, comprising Sol, Terra and Luna, its latest AI models aimed at developers and enterprise users. The launch is currently limited to a small group of trusted partners, with a wider rollout planned in the coming weeks. OpenAI describes Sol as its most capable model yet, designed for advanced reasoning, coding and agentic workflows. The company says it also delivers improvements in scientific research tasks, including biology and cybersecurity.
Alongside Sol, OpenAI is introducing Terra, a balanced model for everyday workloads, and Luna, a lower-cost model optimised for speed. According to the company, Terra delivers performance comparable to GPT-5.5 at roughly half the cost, while Luna is its most affordable offering.
OpenAI is also introducing new capabilities with GPT-5.6. Sol includes a Maximum Reasoning mode for tackling more complex problems and an Ultra mode that uses sub-agents to complete multi-step tasks more efficiently. Beyond software development, the ChatGPT maker says GPT-5.6 performs better than GPT-5.5 on biology research benchmarks while using fewer tokens, and remains competitive in cybersecurity evaluations without crossing its internal "Cyber Critical" safety threshold.
Safety is another major focus of the preview release. OpenAI says GPT-5.6 includes stronger safeguards against cyber abuse, sensitive biological misuse and repeated attempts to bypass its protections. The company says it spent more than 700,000 A100-equivalent GPU hours on automated red-teaming, alongside extensive testing by human experts, to identify vulnerabilities before launch.
GPT-5.6 availability
GPT-5.6 is currently available through the API and Codex for select partners, with broader availability across ChatGPT planned in the coming weeks. OpenAI has priced Sol at $5 per million input tokens and $30 per million output tokens, while Terra costs $2.50 and $15, and Luna $1 and $6, respectively.
