OpenAI is dropping GPT-5.6 Sol’s API input price from five dollars to four dollars per million tokens, and its output price from thirty dollars to twenty dollars per million tokens, a reduction of more than twenty percent. The cut applies to the API, Codex credits and eligible ChatGPT Work plans, runs at least through 21 November 2026, and was posted to OpenAI’s own developer community on 21 August 2026. It follows a DeepSeek rate increase of up to elevenfold depending on model and token type, Anthropic’s cancellation of a planned fifty percent Claude Sonnet 5 price rise, and an eighty percent cut to GPT-5.6 Luna — Sol is the fourth major frontier price move in three weeks, and the first to touch a flagship rather than a mid-tier model.

Why it matters

Sol is OpenAI’s top-tier flagship, the model buyers reach for when they need the best available reasoning rather than the cheapest. Cutting its price rather than holding it as a premium anchor signals that OpenAI is now competing on cost even at the frontier, not just in the mid-tier where Luna already absorbed a cut. Buyers running Sol in production get a direct, time-boxed cost reduction; buyers who chose a cheaper model specifically to avoid Sol’s price should recheck the maths, because Sol’s cost profile has moved.

What happened

OpenAI announced the change through its developer community on 21 August 2026, in a post describing the move as “dropping API and credit pricing of GPT-5.6 Sol by over 20% for the next 3 months.” The specific figures: input tokens fall from 5.00 dollars to 4.00 dollars per million, and output tokens fall from 30.00 dollars to 20.00 dollars per million. That is a twenty percent cut on input and a thirty-three percent cut on output — “over 20%” understates the output-side reduction.

The promotional pricing is explicitly time-boxed. OpenAI’s own post states the reduced rate is available “at least through November 21, 2026,” language that leaves the door open to extending it but does not commit to doing so. This is a promotional rate, not a list-price change — the distinction matters for anyone modelling costs into 2027.

The cut is scoped to specific products: the pay-as-you-go API, Codex credits, and “eligible” ChatGPT Work plans. Consumer subscription tiers — Plus, Pro and Business — are unaffected; their prices stay where they were. This is a developer- and enterprise-facing move, not a consumer one.

The price war context

Frontier AI pricing moved sharply in the space of a week in late July — the eighty percent figure in that piece describes GPT-5.6 Luna’s cut specifically, not a flat market-wide drop, since other vendors moved by smaller and more varied amounts — driven initially by mid-tier and open-weight competition. OpenAI itself had already cut GPT-5.6 Luna’s price by eighty percent, positioning Luna as the volume option. DeepSeek moved in the opposite direction, raising its API rates by as much as elevenfold depending on model, token type and time of day — the elevenfold figure is the top of a range that starts around fifty percent, not a flat rate rise — after a period of undercutting the market, suggesting even the most aggressive discounter needed to recover margin. And Anthropic cancelled a planned price rise for Claude Sonnet 5 rather than let a competitor’s list price sit below its own.

What is different about the Sol cut is the tier. Luna, Sonnet 5 and the DeepSeek adjustment were all mid-tier or volume-model moves — the segment of the market most exposed to open-weight and Chinese-lab competition on cost per token. Sol is OpenAI’s flagship reasoning model, the one typically priced to reflect quality rather than to win a race to the bottom. A twenty-to-thirty percent cut on a flagship, even a temporary one, suggests OpenAI judged that holding Sol’s price steady was costing it enterprise volume that a temporary discount could recover.

Search results reviewed for this story pointed to Google’s Gemini 3.7 Flash launching in the same window “at roughly half the price of the previous generation” and Anthropic “chipping away at enterprise accounts by emphasising reliability and safety tooling” — both claims from secondary aggregator reporting rather than a source in a position to know, and neither is treated as established fact here; they describe the competitive backdrop OpenAI itself is reacting to, not a verified fact about Gemini or Anthropic pricing.

What would settle the trajectory

Two things are still open. First, whether OpenAI extends the promotional rate past 21 November 2026 or lets it lapse back to five and thirty dollars — OpenAI’s own post leaves that undecided. Second, whether this is a one-off adjustment or the start of a broader flagship repricing; OpenAI has not announced changes to Sol’s consumer-tier pricing or to its other flagship-adjacent models.

What this means for buyers

Anyone running production workloads on GPT-5.6 Sol through the API, Codex, or an eligible ChatGPT Work plan gets an automatic cost reduction with no action required — check your plan is on the eligible list before assuming the new rate applies. Anyone who switched away from Sol specifically because of its price should re-run the comparison against Luna, Gemini 3.1 Pro or Claude Opus 5 using the new twenty-and-four figures, since the gap that justified switching may have narrowed. Because the cut is promotional and time-boxed to 21 November 2026, buyers signing longer contracts or building cost models that extend past that date should budget for the possibility of reversion to list price, not assume the discount is permanent.

What would change our reading

If OpenAI confirms the discount becomes the new list price rather than lapsing on 21 November 2026, this becomes a permanent flagship repricing rather than a promotion, which is a stronger signal about margin pressure at the top of the market. If OpenAI extends the same treatment to consumer Pro and Business tiers, that would indicate the move is about market share broadly rather than developer retention specifically. Conversely, if the rate lapses back to five and thirty dollars in November without comment, that would suggest this was a short-term response to a specific competitive event rather than a structural repricing.

Sources

  • OpenAI Developer Community, “20% price reduction for GPT 5.6 Sol: API, Codex credits and ChatGPT Work,” 21 August 2026 — community.openai.com
  • Business Standard, “OpenAI cuts developer pricing for GPT-5.6 Sol model by more than 20%,” 22 August 2026 — business-standard.com
  • citybiz, “OpenAI Cuts GPT-5.6 Sol API Pricing 20% for Three Months” — citybiz.co