OpenAI cuts GPT-5.6 Sol by more than 20 percent for three months

OpenAI is cutting the price of its frontier model GPT-5.6 Sol. Not as a new list price. As a time-boxed reduction of more than 20 percent on the API and on credits for ChatGPT Work and Codex, available at least through 21 November 2026. Plus, Pro and Business subscription allowances stay the same.

That distinction matters for CIOs and CFOs. Sol is not a chat model you toggle for a trial. It is the engine inside coding agents, long-horizon workflows and the tasks where a wrong answer costs more than the tokens. Cheaper Sol for three months does not change capability. It changes the temptation to pour more volume onto the top model before anyone owns routing and caps.

What actually moved

OpenAI announced the cut on 21 August in its developer forum and on the company’s official X account. Reuters confirmed the same evening and framed it as a response to competition from Anthropic and Chinese models.

For short-context calls, still the bulk of production traffic, input falls from $5 to $4 per million tokens and output from $30 to $20. Cached input falls from $0.50 to $0.40. That is 20 percent on input and about 33 percent on output. Taken together it is more than 20 percent, which is how OpenAI wrote it.

The live pricing table also shows the new long-context Sol list: $8 input, $0.80 cached input and $30 output per million tokens. Regional processing with data residency still carries a 10 percent uplift for models released on or after 5 March 2026.

The cut is live on the API and rolling out across eligible ChatGPT Work and Codex credit plans. OpenAI is explicit about what does not change: included usage on Plus, Pro and Business. You do not get more Sol inside the seat license. You get a lower unit price when you pay per token or per credit.

This is the third GPT-5.6 price move in three weeks. On 30 July OpenAI cut Luna by 80 percent and Terra by 20 percent, and launched Fast mode for Sol: up to 2.5 times Standard speed at twice the price. Sol’s list price stayed put then. Now Sol follows, but as a promotion, not a new floor.

Why the output cut is the one that moves the bill

In plain Q&A, input often dominates. In agents it is the other way around. A coding agent plans, reads a repo, calls tools, writes a patch, runs tests and tries again. Output and intermediate steps eat the budget. A drop from $30 to $20 per million output tokens is not cosmetic. It is the line item that made it expensive to let Sol run without a cap.

Reuters puts the new Sol price next to Anthropic: Claude Fable 5 is listed at $10 input and $50 output, Claude Opus 5 at $5 and $25. Sol at $4 and $20 is cheaper than Opus 5 on paper. That is a procurement talking point. It is not a governance argument. Fable, Opus and Sol are not interchangeable on cyber behavior, refusals, data handling or how they fail inside a pull request. Price alone is a weak SLA.

The cyber track is untouched. On OpenAI’s pricing page, gpt-5.6-cyber is $12.50 input and $75 output per million tokens at short context. Daybreak aliases still point at Sol and the cyber model. This promotion does not make the most expensive security capacity cheap.

This is a window, not a new normal

21 November is not a footnote. If your agent fleet is rerouted onto Sol because it is “now cheap”, and the promotion expires without a Terra and Luna plan, you will see a unit-cost jump in the fourth quarter. That is standard vendor management: temporary price is used to win default routing.

hogby.ai wrote on 30 July that OpenAI was moving AI control from model choice to cost per outcome. That still holds. The difference now is that the flagship is inside the price war, but only on a clock. Luna and Terra got lasting cuts. Sol got a discount window. Budget them differently.

In the same week OpenAI said it is slowing parts of frontier reinforcement learning while security catches up with capability, after a test agent reached Hugging Face. A price cut and a safety brake can coexist. One is about winning volume. The other is about capabilities already dangerous enough that the lab itself slowed down. For a board the point is simple: cheaper Sol means more Sol volume, not less need for logging, sandboxes and human approval.

What leaders should do before 21 November

Set caps per team and per agent, not only per API key. Sol as the default in Codex or ChatGPT Work will look rational in August and expensive in December if the promotion lapses.

Route by cost of being wrong. Sol for requirements, planning and work where a bad patch costs more than tokens. Terra for everyday production. Luna for volume, auto-review and tightly specified steps. Fast mode only where latency has a price you can name.

Require a record of which model actually ran. If auto-routing sends more work to Sol because it is temporarily cheaper, you lose the July baseline.

Do not read unchanged Plus, Pro and Business plans as “no effect”. The seat license is the same. The credit and API bill is what moves. Where agents already consume the token budget, the cut will show. Where people sit in chat on included quota, it barely will.

And remember that OpenAI on Amazon Bedrock is billed through AWS and may differ. Organizations with data residency, an EU region or a hyperscaler contract need the actual invoice line, not only openai.com.

Sources and media

Primary source: OpenAI, “20% price reduction for GPT 5.6 Sol: API, Codex credits and ChatGPT Work”, 21 August 2026: https://community.openai.com/t/20-price-reduction-for-gpt-5-6-sol-api-codex-credits-and-chatgpt-work/1391726

Confirmed price list: OpenAI API Pricing, with promotional language at least through 21 November 2026: https://developers.openai.com/api/docs/pricing

Official announcement on X: @OpenAI, 21 August 2026: https://x.com/OpenAI/status/2090885187634905500

Corroboration: Reuters, “OpenAI cuts developer pricing for frontier GPT-5.6 Sol model by more than 20%”, 21 August 2026 21:29 UTC: https://www.reuters.com/technology/openai-cuts-developer-pricing-frontier-gpt-56-sol-model-by-more-than-20-2026-08-21/

Earlier cut in the same family: OpenAI, “Advancing the price-performance frontier with GPT-5.6”, 30 July 2026: https://openai.com/index/advancing-the-price-performance-frontier-with-gpt-5-6/

Business pricing showing Sol at $4/$20 after the cut: https://openai.com/business/pricing/

Thumbnail: OpenAI Image 2 / hogby.ai