GPT-6.1 Sol

GPT-6.1 Sol arrived on September 29, 2026 at OpenAI's DevDay as gpt-6.1-sol, an upgrade to GPT-6 Sol that OpenAI pitches as near-Astra capability at one fifth of Astra's token prices. The rate is $2 per million input tokens and $10 output, unchanged from GPT-6 Sol, with cached input cut in half to $0.10 and cache writes at $2.50. The long-context rule carries over: prompts above 272,000 input tokens bill at $4 input, $0.20 cached, and $15 output. The window is 1,050,000 tokens with 128K output, text and image input, a knowledge cutoff of April 30, 2026, and reasoning effort from low through max with medium as the default. It is in the API and in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users; OpenAI says it is not yet in regular Chat, and that an Ultrafast tier for it will follow in the days after launch. OpenAI's launch claims are relative rather than tabulated: it matches GPT-6 Astra on DeepSWE v1.1 at roughly one fifth of the cost and beats GPT-6 Sol's best score there by 6.4 points, scores 2.2 points above Claude Opus 5.5 on AutomationBench at medium effort, comes within 2.1 points of Astra on the offline OSWorld 2.0 set at about one seventh of the cost per task, and cuts the share of answers containing a factual error at low effort from 11.4 percent to 7.7. The independent numbers back the direction. Vals measures Terminal-Bench 4.0 at 55.05 percent on its mini-swe-agent run, against 44.44 for GPT-6 Sol and 59.60 for Astra, and places it eighth on the Vals Index at 61.15 percent as of September 30, 2026. Cognition scores it 50.2 percent on FrontierCode v1.1 at medium effort, ahead of GPT-6 Sol at 49.3, and Artificial Analysis scores it 52, one point behind Astra. OpenAI's models page now features Astra, GPT-6.1 Sol, and Luna; GPT-6 Sol is still on the price list and is not deprecated. It lands at the same $2 and $10 as Claude Sonnet 5.5, released a day earlier, which is the comparison most buyers will run.