OpenAI launched GPT-6.1 Sol during its DevDay keynote Tuesday, pitching it as a lower cost model that nearly matches the capability of its flagship GPT-6 Astra, according to OpenAI's announcement and TechCrunch. Sol costs one-fifth of Astra's standard input and output token prices, both outlets reported.
OpenAI priced Sol at $2 per million input tokens and $10 per million output tokens, with cached input discounted 95% to $0.10, according to a summary of the pricing published by Latent Space. A faster ultrafast mode runs at six times that price, comparable to Astra's own $60 and $300 per million token rates, Latent Space reported.
Independent testing from Artificial Analysis found Sol scored one point below Astra on its Intelligence Index while costing $0.72 per task versus $3.26 for Astra, according to Latent Space, which aggregated the benchmark. In a coding test cited in the same report, Sol found 44 bugs for $6.56 compared with 45 bugs for $33 using Astra.
OpenAI also said Sol's factual accuracy improved over its predecessor, GPT-6 Sol, with error rates falling from 11.4% to 7.7% at low reasoning effort, TechCrunch reported. The model became available Tuesday to ChatGPT Plus, Pro, Business, Enterprise and Edu subscribers, currently through ChatGPT Work and Codex, according to TechCrunch.
For developers, the number that matters is the near miss on quality at a fraction of the price. A model that lands within a point of the flagship on capability while costing roughly five times less changes the math on which tasks get routed to the expensive model at all.