OpenAI's GPT-5.6 Sol and ChatGPT Work take on Anthropic with lower cost and speed

OpenAI today announced GPT-5.6 Sol, Terra, and Luna — three new models aimed squarely at Anthropic's Fable 5 and Mythos 5. The flagship Sol model scored 53.6 on Agents' Last Exam, beating Fable 5 by 13.1 points. On the Artificial Analysis Intelligence Index, Sol came within one point of Fable 5 while using 61% less time. In coding benchmarks, Sol reached 80 on the AI Coding Agent Index, 2.8 points ahead of Fable 5. At medium reasoning, Sol outperforms Fable by 11.4 points at roughly one-quarter the estimated cost, according to OpenAI. Qodo's CEO noted Sol uses 3x fewer tokens per PR with 2x lower latency than GPT-5.5. The models include enhanced safeguards combining model protections and real-time monitoring.

Separately, ChatGPT Work debuts as OpenAI's agentic desktop productivity tool, clearly an answer to Anthropic's Claude Cowork. Previously embedded in Codex (OpenAI's coding agent), ChatGPT Work now allows non-coders to let the AI access desktop and browser to perform tasks. This move shows improved branding strategy from OpenAI. The launch comes after government intervention delayed GPT-5.6's release last month and amid intense industry buzz around Anthropic's recent Fable 5 announcements. OpenAI is positioning these releases as a direct competitive response, offering faster, cheaper, and more accessible AI agents.