OpenAI cuts GPT-5.6 Sol API rates to $4 input and $20 output

• GPT-5.6 Sol's three-month promotion puts API input at $4 and output at $20 per million tokens. OpenAI called it a reduction of more than 20% in its pricing announcement, while Amazon Bedrock's August 21 notice gives the exact rates.
• OpenAI's public API pricing page still shows the prior $5 input and $30 output standard schedule, the same before-and-after rates haider1 cited in a pricing post.
• API access is live, while eligible ChatGPT Work and Codex credit plans are rolling out, according to OpenAIDevs' rollout post. Pro, Plus, and Business subscription usage remains unchanged, OpenAI says in its announcement.
• On Devin Desktop and CLI, cognition says its existing 70% promotion stacks with OpenAI's new cut to reach 76% off list price through October 3 in cognition's announcement.

The API pricing page labels its $5 and $30 figures as standard processing below 270K context tokens and separately lists cached input at $0.50 per million. OpenAI also paired the rate news with API-key attribution and spend limits, while cognition published a FrontierCode price-performance plot alongside its Devin-specific promotion.

OpenAI says the promotion lasts three months. Amazon Bedrock's corresponding announcement says it is available at least through November 21.

• Input tokens: $5 to $4 per million, a 20% reduction.
• Output tokens: $30 to $20 per million, a 33.3% reduction.

Output billing gets the larger cut, despite OpenAI framing the announcement as an over-20% reduction.

At publication, OpenAI's API pricing page still listed GPT-5.6 Sol at $5 per million input tokens, $0.50 cached input, and $30 output for standard processing below 270K tokens. Its announcement gives the promotional input and output figures, but no cached-input price or long-context schedule.

The two official surfaces therefore present both the standard $5/$30 schedule and the temporary $4/$20 promotion.

OpenAI said the lower rate is live for API users, with rollout across eligible ChatGPT Work and Codex credit plans continuing through the day.

• API Platform: available now.
• ChatGPT Work: eligible credit plans are rolling out.
• Codex: credit pricing is rolling out.
• Pro, Plus, Business: included subscription usage remains unchanged.

OpenAI called out eligibility but did not enumerate the covered Work or Codex plans in OpenAIDevs' rollout post.

cognition says OpenAI's 20% reduction stacks with a 70% promotion it had already applied to Sol on Devin Desktop and CLI. The sequential discounts produce the stated 76% reduction from list price, which cognition says runs through October 3.

The company calls Sol its most affordable frontier model on those surfaces in cognition's announcement.

haider1 attached a FrontierCode 1.1 Extended table that places GPT-5.6 Sol at 73% Pass@1 and $8.39 average cost per rollout in its max setting. The same table lists 60K output tokens and 61 steps, against Claude Opus 5's 74%, $11.84, 118K output tokens, and 99 steps in haider1's chart.

Those dollars are workload-level average costs, alongside Pass@1, output-token, and step counts, rather than the published per-million-token API schedule.

OpenAI also added per-key usage and spend tracking to its dashboards, so organizations can attribute spend to individual applications or workloads. The update supports monthly limits at the organization or project level, including hard limits that stop traffic when reached.

The controls are available in the API Platform and through the Admin API for programmatic workflows, according to OpenAIDevs' dashboard update.