AI pricing moved decisively this week. On August 21, 2026, OpenAI announced a promotional price reduction on GPT-5.6 Sol, its most capable model — slashing output token costs by 33% and input by 20%. The new rates: $4 per million input tokens (down from $5), $20 per million output tokens (down from $30), and $0.40 per million for cached inputs (down from $0.50).
The cuts apply to the pay-as-you-go API, Codex credits, and ChatGPT Work plans — not consumer subscriptions — and are valid through at least November 21, 2026.
The Competitive Context
This is OpenAI's third pricing adjustment in roughly a month. Terra and Luna prices came down in late July; Sol — the flagship — held firm until now. The timing is pointed: until August 21, Anthropic's Claude Opus 5 cost less than Sol on both input and output. With the new rates, Sol now undercuts Opus 5 on both dimensions simultaneously — a reversal that signals OpenAI's intent to own the developer segment as the frontier model race enters a new phase.
GPT-5.6 is a family of three models launched on July 9, 2026: Luna (lightweight, free for all ChatGPT users), Terra (midrange), and Sol (flagship reasoning and long-context tasks). Sol has been the go-to for agentic pipelines and multi-step code generation — workloads that generate substantial output token volume.
What the Math Means for Builders
The output token drop hits hardest where many production AI applications spend the most: on generation, not on reading context. A system processing 10 million API calls per day with average outputs of 500 tokens saves roughly $50,000 per month at the new rates compared to last week's pricing. For teams running continuous code agents, legal document drafting pipelines, or high-volume customer service automation, that difference is material.
There's also a quality-of-life update to note: on August 19, OpenAI added an "effort slider" to GPT-5.6 Sol in ChatGPT, letting Plus and Pro users tune how much reasoning the model invests per response. The slider pairs naturally with the API price cuts — developers can now dial between cost and quality at the application level.
The Broader Pricing War
The promotional window through November 21 coincides with what market observers expect to be OpenAI's IPO roadshow window. Cost competitiveness is a key narrative — particularly as Anthropic's own IPO filings have attracted significant investor attention in recent months.
For developers making platform decisions right now, the competitive landscape has materially shifted: frontier-grade reasoning on Sol is no longer priced at a premium over its closest competitor. Whether that holds after November depends on how the market responds between now and then.