What actually happened
OpenAI released GPT-6.1 Sol on 29 September 2026, a week after GPT-6 Sol. API id: gpt-6.1-sol. Standard price is unchanged at $2 per million input tokens and $10 per million output tokens. Cached input is now $0.10, half of GPT-6 Sol’s $0.20 and 95 percent below standard input. Astra remains $10 / $50, with $1 cache.
The model is live for Plus, Pro, Business, Enterprise and Edu in ChatGPT Work and Codex, and for all API users. It is not in Chat. Ultrafast for 6.1 Sol is promised in the coming days. OpenAI’s own evals put it near Astra on DeepSWE coding, GDP.pdf professional documents, Zapier’s AutomationBench and OSWorld computer use, at a fraction of Astra’s cost per task. Those are vendor benches. Treat them as direction, not a guarantee on your stack.
Why a small team cares
Agent bills are cache bills. If a Make scenario or a Codex loop resends the same brief, tickets and schema on every turn, the $0.10 cache line is the cut you feel. Same sticker as last week, better work, cheaper reuse. That is enough to change the default this week without touching Astra for the ugly jobs.
A solo running a nightly reporting agent, or an agency pointing n8n at client docs, should not still be on GPT-6 Sol or a 5.x leftover.
Hype vs useful
“One-fifth the price of Astra” is the standard token ratio, not a promise that every task costs one-fifth. Long prompts above the published threshold pay more. “Near-Astra” is OpenAI’s claim on selected benches. Use Sol as the workhorse. Keep Astra where Sol already fails. Do not rebuild the orchestration layer for a model bump.
