What OpenAI announced
OpenAI cut prices on its GPT-5.6 Sol API by more than 20 percent for three months, the company said in a community pricing post published August 21, 2026. Input tokens fall from $5 to $4 per million, output tokens from $30 to $20 per million, and cached input from $0.50 to $0.40 per million. The lower rates also cover eligible ChatGPT Work and Codex credit plans, while Pro, Plus, and Business subscription usage keeps its existing price. OpenAI set the promotion to run through November 21, 2026, and the post makes no commitment about what the rate does after that date.
The price change in one table
The cut applies to three separate rates, and each one falls by a different amount.
| Token type | Price before | Price now |
|---|---|---|
| Input | $5.00 per million tokens | $4.00 per million tokens |
| Output | $30.00 per million tokens | $20.00 per million tokens |
| Cached input | $0.50 per million tokens | $0.40 per million tokens |
Why now: competitors set the price
Reuters and Yahoo Finance both reported OpenAI's price cut as a direct response to competition from Anthropic and from Chinese model vendors, including DeepSeek. BigGo Finance corroborated the same framing on the same day, August 21, 2026, describing the move as OpenAI defending Sol's position in a market where rival labs keep undercutting each other on price. None of the coverage points to a drop in OpenAI's own running costs as the reason for the cut. That distinction matters for anyone reading the discount as a sign that AI inference has simply gotten cheaper to produce, because the reporting says otherwise.
Why an EU or UK owner should not build on this rate
A three-month, competitor-driven discount is not a stable price floor for any business signing a new API contract this quarter. The rate exists because OpenAI is reacting to Anthropic and to Chinese vendors, not because its own costs fell, so the ordinary case is that pricing reverts toward the old $5/$30/$0.50 numbers once the promotion ends on November 21, 2026, and the competitive picture could just as easily push it higher. An owner who is signing a new contract or scaling usage during this window should renegotiate terms or lock a rate now, rather than building a twelve-month cost model on three months of discounted tokens. For a business billed in euros or pounds, the exposure is doubled, since the dollar price itself can move again on top of the exchange rate. This is the same pattern Servola flagged this week in DeepSeek's own pricing, where a headline rate turned out to hide a peak-hour surcharge: across vendors, the number on the announcement is proving to be a temporary lever, not a structural cost, and a budget should be built for the reversal, not the promotion.
Read next: 1 Billion Weekly Users, Free ChatGPT Goes Unlimited | OpenAI's Mid Tier Now Costs Ten Times Its Cheapest
