Fahad Manzur
OpenAI released GPT-6.1 Sol at its DevDay event on September 29, pricing it at one-fifth of what GPT-6 Astra costs per token while nearly matching Astra on agentic coding, computer use, and professional-work benchmarks. Standard API pricing for Sol is $2 per million input tokens and $10 per million output tokens, against Astra’s $10 and $50.
What changed under the hood
Per OpenAI’s own announcement, Sol arrived one week after the base GPT-6 Sol release and 26 days after Astra, and at low reasoning effort its factual-error rate on OpenAI’s internal evaluation dropped from 11.4 percent to 7.7 percent. The model is available now inside ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu accounts, though it has not yet rolled out to the regular ChatGPT consumer tier. A faster variant, Sol Ultrafast, is coming within days and promises up to eight times quicker token generation inside Codex.
Why the price cut is the real story
Model-to-model benchmark gains get smaller every release; the pricing gap this time is not. TechCrunch’s coverage of the launch frames Sol as part of a pattern where OpenAI ships a cheaper near-equivalent to its own flagship model within weeks, rather than waiting for a competitor to force the price down. For anyone running an AI-powered service, a chatbot, an automated research tool, a coding assistant wrapped into a product, the cost of a million output tokens just fell by five times without giving up much capability.
What it means for people building on top of these models
A freelance developer or small agency charging clients for AI-assisted work has historically had to eat Astra-level API costs or settle for a noticeably weaker, cheaper model. Sol narrows that trade-off: near-Astra output quality at a fraction of the bill. That matters more in markets where clients pay in dollars but the margin on a project gets calculated in rupees or dirhams, where a few cents per thousand tokens is the difference between a profitable retainer and a break-even one. The catch is availability: Sol is locked to ChatGPT Work, Codex, and the API for now, so anyone building through the consumer ChatGPT app will have to wait for a wider rollout before they can switch.
