GPT-6.1 Sol: The Most Aggressive Price Cut of the Year
September 30, 2026 · 5 min read
For the last two years, every model lab has followed the same script: launch an expensive flagship, then six months later ship a cheaper "lite" version that's 30% less capable and 20% cheaper. OpenAI just tore up the script.
At its DevDay on September 30, OpenAI unveiled GPT-6.1 Sol with a one-line pitch: near-flagship GPT-6 Astra capability at one-fifth the price. Standard API input and output are both priced at a fifth of Astra's rates; cached input is even more dramatic, at one-tenth. OpenAI is calling it, without blushing, "the best value model at its performance level."
Who actually benefits
It's already live where it matters: ChatGPT Work, Codex, and the API — and the positioning is explicit: agentic coding, computer use, and professional work. Those are exactly the workloads that burn the most tokens. For a team that has AI writing code all day, cutting the model bill to a fifth isn't a rounding error; over a year it can be the difference of tens of thousands of dollars.
Three things to check before you switch
- First, "near Astra" is OpenAI's claim. How much is actually lost in the shrink needs independent benchmarks, not a keynote slide.
- Second, the Ultrafast variant isn't out yet. It's promised in the coming days — if you need maximum speed, wait for that one.
- Third, the price cut is, at bottom, about competition. Anthropic just closed a funding round at a nosebleed valuation, and DeepSeek's V4.1 Flash set a new bar for speed. OpenAI couldn't afford not to move.
The takeaway
For most people the conclusion is simple: stop defaulting to the most expensive flagship. Try Sol first and see if it holds up. Money should go where it matters — and now, so should your model choice.