OpenAI cuts GPT-5.6 Sol list prices: input down 20%, output down a third

OpenAI has lowered the list price of GPT-5.6 Sol, reducing input token rates by 20% and output token rates by roughly a third. According to the Vercel blog, the cut applies to rates billed per million tokens for requests up to 272K tokens, and because the model ID is unchanged, requests developers already send are billed at the new price automatically.

What the price change covers

Vercel outlines the scope of the reduction:

• Input tokens cost 20% less than the previous list price.
• Output tokens are down by about a third.
• Cached tokens, cache writes, long-context requests above 272K tokens, and US regional rates all drop by the same proportion as the headline cut.
• The discount applies across every OpenAI service tier.

Vercel points readers to its pricing page for the full rate tables, since the changelog lists proportions rather than absolute per-token dollar figures.

How the gateway discount stacks

Vercel's AI Gateway, which routes requests to model providers, runs a 50% discount on OpenAI models, and that discount now applies to the new lower list price through September 18. Because the gateway reduction is a percentage rather than a fixed deduction, the two cuts compound: traffic routed through the gateway is billed at half of an already-reduced rate for the remainder of the promotional window.

No code changes required

The model identifier stays openai/gpt-5.6-sol, so integrations that already target the model pick up the new pricing without modification. Vercel shows the same AI SDK streamText call continuing to work, just at a cheaper rate. One exception the company flags: bring-your-own-key requests bill at your own rate with OpenAI, so those see OpenAI's list-price cut but not the gateway's half-off.

Using Sol in coding agents

Vercel highlights Sol for agentic development workflows, with an example that has the model investigate failing tests and open a pull request with a fix. To connect coding agents to the gateway, developers can run vercel ai-gateway coding-agents setup and then select openai/gpt-5.6-sol in the agent's model settings. The model is also available in Vercel's playground alongside the rest of the gateway's model catalog.

Why it matters

Agentic workloads tend to be output-heavy, built from long chains of tool calls, reasoning traces, and generated code, so a one-third cut on output tokens is where most of the savings will concentrate for developers running coding agents on Sol. The proportional reduction across cached tokens and long-context rates also benefits retrieval-augmented and large-context pipelines, which lean on both. The change also illustrates how platform gateways pass vendor price moves through: when a discount is keyed to list price, a provider-side cut flows straight to the customer without any reconfiguration. The date to watch is September 18, when the gateway's 50% promotion window closes and routed traffic reverts to the new list price unless the offer is extended.