OpenAI posted on its developer forum at 19:41 UTC on 21 August that it is "dropping API and credit pricing of GPT-5.6 Sol by over 20% for the next 3 months." The new rates are already live on the company's pricing page: standard short-context $4.00 input, $0.40 cached input and $20.00 output per million tokens, against $5.00, $0.50 and $30.00 before. Long-context rates sit separately at $8.00 and $30.00.
The 20% is the smallest of the three cuts
Input fell exactly 20% and cached input exactly 20%. Output fell 33%, from $30 to $20. Agentic and coding workloads are output-heavy by construction — a Codex-style loop emits far more tokens than it ingests — so the blended saving for the workloads OpenAI is courting is materially larger than the number in every headline. "More than 20%" is true and misleading in the cheap direction.
What the common framing gets wrong
Every write-up says OpenAI "cut prices." OpenAI said it is dropping pricing for the next three months. A reply in the same announcement thread states the promotional rate runs at least through 21 November 2026. Anyone modelling unit economics on $4/$20 is modelling a coupon with an expiry, not a structural decline in inference cost. The second misreading is who benefits: the change reaches the API and credit balances on ChatGPT Work and Codex. OpenAI's own sentence is unambiguous — "Pro, Plus, and Business subscription usage remains unchanged." That is where nearly all the users are, and their bill did not move.
The stated reason is not the reported one
Coverage attributes the move to competitive pressure. OpenAI's stated reason is efficiency: "as we continue to push the frontier of capabilities while improving efficiency." It is the same reason given in July, when GPT-5.6 Terra fell 20% and GPT-5.6 Luna fell 80%. Three models in the same family repriced inside a month says more about the economics of the 5.6 generation than about any rival's launch calendar — though a promotion, unlike a price cut, is what a company runs when it wants the volume without conceding the rate.
What to watch on 21 November
The tell will be whether the rate reverts, is extended, or quietly becomes list. Promotions that lapse are pricing experiments; promotions that stick were price cuts that the company was not ready to call permanent.
