OpenAI Cuts GPT-5.6 Sol API Pricing by Over 20% — Now Undercutting Claude Opus 5
GPT-5.6 Sol drops to $4 per million input tokens and $20 per million output tokens, down from $5 and $30. The promotional pricing runs three months and lands the same week Google cut Gemini 3.7 Flash prices.
OpenAI cut API pricing for GPT-5.6 Sol, its flagship frontier model, by more than 20% on August 21. Input tokens now cost $4 per million, down from $5. Output tokens fall harder: $20 per million, down from $30 — a 33% reduction on the side of the meter that dominates most agentic workloads.
The new rates apply to standard short-context use on the API and roll out across eligible credit plans for ChatGPT Work and Codex, OpenAI’s coding tool. Pro, Plus, and Business subscriptions stay at current prices. The cut is promotional, valid for three months through November 21 — which gives OpenAI a built-in option to quietly extend it or let it lapse depending on what competitors do next.
That competitive context is the real story. This is the second pricing move on the 5.6 family in a month, and it lands in the middle of an industry-wide repricing wave: Google cut Gemini 3.7 Flash to $0.75 and $3.75 per million tokens the same week, while DeepSeek moved in the opposite direction and raised V4-Pro’s peak-hour output pricing to $3.96 per million. Frontier inference is turning into a knife fight on price at exactly the moment Anthropic is courting public-market investors ahead of an IPO.
The cut also puts GPT-5.6 Sol below Claude Opus 5 on list price — a positioning move aimed squarely at developers deciding which frontier model anchors their production stack. Token prices are now a marketing instrument, not just a cost line.
For developers, the math is straightforward. A workload burning 100 million output tokens a month on Sol just went from $3,000 to $2,000. Agentic systems with heavy tool-call loops — where output tokens routinely outnumber input — benefit most. If you deferred moving a batch pipeline or evaluation suite to a frontier model because of cost, the three-month window is the time to benchmark it.
Two caveats. First, promotional pricing is not committed pricing: budget against the November 21 expiry, not the current rate. Second, list price is a poor proxy for effective cost — cached input, batch discounts, and context-length tiers move real bills far more than a 20% headline cut. Run your own traces before switching.
Still, the direction is unmistakable. Eighteen months ago frontier-model pricing only went up between releases. Now the top labs are cutting mid-cycle to defend developer share. Whatever else happens this quarter, inference is getting cheaper — and the labs are absorbing the margin hit to keep you on their stack.