OpenAI Lowers GPT-5.6 Sol API and Token Rates by Over 20 Percent

OpenAI reduced developer API and credit pricing for its GPT-5.6 Sol model by over 20 percent on August 22, 2026. The price cut lowers input costs to $4 per million tokens and output costs to $20 per million tokens while preserving existing prompt caching discounts. The promotional rates apply across developer API requests, ChatGPT Work, and Codex credit usage through November 21, 2026, while individual subscription tiers remain unaffected.

The New Per-Token Price Structure for GPT-5.6 Sol

The updated rate schedule reduces both sides of the standard token billing model for the GPT-5.6 Sol model.

Under the revised pricing, standard input drops from $5 to $4 per million tokens, representing a 20 percent discount on incoming text and instructions processed by the model.

Output pricing drops from $30 to $20 per million tokens, a 33.3 percent reduction on all model-generated responses.

The structure of the price adjustments breaks down as follows:

• Standard input tokens: $4 per million tokens (down from $5)
• Standard output tokens: $20 per million tokens (down from $30)
• Cached input tokens: Maintain a tenfold discount relative to standard inputs

Prompt caching allows developers to reuse repetitive context—such as system prompts, extensive documentation, or static codebases—without paying full price on recurring calls. By keeping the tenfold discount in place, cached input queries are billed at one-tenth of the new base rate for the duration of the promotion.

How API and Credit-Based Pricing Functions

In large language models, compute usage is metered in tokens rather than raw character counts or clock time. A token is a sequence of characters that roughly corresponds to three-quarters of a standard English word.

When a developer submits a request to an API endpoint, the request splits into two distinct operational phases:

1. Input processing: The model reads, parses, and encodes the prompt and any attached background documents.
2. Output generation: The model generates subsequent tokens sequentially, using compute cycles for each generated unit.

Output generation requires significantly more computational resources than reading input text, which is why model providers historically set output token rates higher than input rates.

When developers send identical context blocks across multiple API requests, prompt caching systems store the pre-computed representations of those static inputs in memory. This skips the initial parsing step on subsequent calls, lowering compute overhead on the provider’s infrastructure and reducing billing costs for the developer.

Eligible Workloads and Term Dates

The discounted rates take effect immediately and remain active through November 21, 2026.

The temporary rate cut applies to standard API calls as well as credit consumption within ChatGPT Work and Codex environments. These enterprise and developer workspaces meter usage against shared organizational credit pools that draw directly against backend model runtimes.

The rate adjustments do not apply to flat-rate consumer subscriptions. Fixed-price tiers, including ChatGPT Pro and Plus accounts, operate under separate monthly subscription billing and do not receive metered credit adjustments under this change.