GPT-6.1 Sol

GPT-6.1 Sol is the model OpenAI released on 29 September 2026 as gpt-6.1-sol on the Responses and Chat Completions APIs, according to its API changelog of that day, which also lists multi-agent delegation to sub-agents as a beta feature of the model. OpenAI’s model guide now recommends GPT-6 Astra for the hardest reasoning and coding, GPT-6.1 Sol “to balance intelligence and cost”, and GPT-6 Luna for high-volume work. The model page gives a context window of 1.05 million tokens, up to 128,000 output tokens, an April 2026 knowledge cutoff, text and image input with text output, five reasoning-effort settings from low to max, and support for web and file search, code interpreter, computer use, MCP and other hosted tools; US and EU data residency are offered, though the faster tier is not available with EU residency. The standard price is $2 per million input tokens and $10 per million output tokens, with cached input at $0.10; long-context requests cost $4 and $15, batch and flex processing half the standard rate, and a faster tier twice it — against $10 and $50 for GPT-6 Astra, so about a fifth of Astra’s price, and the same list price as GPT-6 Sol and Claude Sonnet 5.5. The Decoder, reporting the release independently the same day, says Sol is available to paying ChatGPT Work and Codex users and in the API, and that OpenAI has held back its planned flagship, GPT-6.1 Astra, over safety concerns. OpenAI’s own benchmarks, which it calls preliminary — a tie with Astra on DeepSWE v1.1, 2.1 points behind it on OSWorld 2.0, and ahead of Claude Opus 5.5 on GDP.pdf and AutomationBench — are its own evaluation and are not stored as scores. Artificial Analysis, measuring independently, scores it 52 on version 4.3.2 of its Intelligence Index at maximum effort, against 56 for Claude Sonnet 5.5 and 58 for Claude Opus 5.5 in the same version, and records 67 million output tokens to complete the index.