On August 21, OpenAI cut the API prices for GPT-5.6 Sol — by more than 20% for the next three months. Instead of $5 you now pay $4 per million input tokens, and output drops from $30 to $20. That’s 20% off input and a third off output.
Who it affects — and who it doesn’t
The cut applies to the API and to eligible credits on ChatGPT Work and Codex. The Pro, Plus, and Business subscriptions stay the same. So if you work through the interface, you won’t notice a thing. If you build on the API or Codex, you’re saving from now on. The promo runs at least through November 21.
The pattern behind it is the interesting part. This is already the second pricing round on the 5.6 family within a month — in late July, OpenAI had already cut prices on its Terra and Luna models. This isn’t a one-off adjustment, it’s a steady drumbeat.
The message to Anthropic
Several observers read the cut as a direct counter to Claude Opus 5, which has been on the market since late July. And the timing fits: just a few days ago, Anthropic walked back an announced price increase for Sonnet 5. Both sides are feeling out who can afford which price for which performance.
For those of us building with these models, that’s good news for now. The competition at the top isn’t just about benchmarks anymore — it’s about price per token. And token prices are what shows up on the invoice at the end of the month.
My take
Three months of discount isn’t a new list price — it’s a promo with an expiry date. The question is what happens in November. If the price stays down, it was a real cut. If it snaps back, it was marketing.
Still: two years ago, nobody would have expected a frontier model to compete this openly on price this fast. Compute is getting cheaper, the models are fighting harder, and for now the winners are the people who build with them. I’m watching the numbers closely — but I’m not blindly shifting my pipelines over it. You don’t switch a model over three months of discount.
Sources: