OpenAI lists GPT-5.6 Sol at $4 per 1 million standard short-context input tokens and $20 per 1 million output tokens. That is the direct answer for developers budgeting ordinary API requests—but it is not the whole bill: cached tokens, cache writes and long-context requests use separate rates.
GPT-5.6 Sol API rates at a glance
A token is a small unit of text processed by an AI model. API billing counts the tokens sent to the model as input and the tokens it generates as output. For standard short-context requests, the listed rates are:
- Input: $4 per 1 million tokens
- Output: $20 per 1 million tokens
The input and output prices matter separately. An application that sends large prompts but generates short replies will feel the input rate more strongly; a coding or content workflow that produces long responses will spend more on output.
Every rate that can change the bill
Cached input is text the system can reuse instead of processing as a new full input. A cache write is the charge for storing that reusable context. Those categories are not interchangeable with ordinary input or output tokens.
| Request category | Input | Cached input | Cache writes | Output |
| Standard short context | $4 / 1M tokens | $0.40 / 1M tokens | $5 / 1M tokens | $20 / 1M tokens |
| Long context | $8 / 1M tokens | $0.80 / 1M tokens | $10 / 1M tokens | $30 / 1M tokens |
The table’s most important warning is simple: do not apply the $4/$20 pair to every request. Long-context work has its own prices, and cached input has a lower read rate while cache writes carry a different charge.
Short context and long context are different prices
The $4 input and $20 output rates apply to standard short-context requests. OpenAI lists long-context GPT-5.6 Sol usage at $8 per 1 million input tokens and $30 per 1 million output tokens, with cached input at $0.80 and cache writes at $10 per 1 million tokens.
That distinction matters when an application sends large documents, extensive conversation history or other unusually long prompts. A developer estimating costs with the standard rates could understate the bill if the request falls into the long-context category.
How much cheaper is GPT-5.6 Sol now?
The reported earlier standard rates were $5 per 1 million input tokens and $30 per 1 million output tokens. Against those figures, the input rate fell by 20%, from $5 to $4, while the output rate fell by approximately 33.3%, from $30 to $20.
Those are two different reductions, not one universal discount applied to every workload. The effective saving depends on the ratio of input to output tokens. For example, a prompt-heavy application benefits more from the input reduction, while an application that generates long responses benefits more from the output reduction.
The practical rule is to estimate both sides of your own traffic:
- Count the input tokens your application sends.
- Count the output tokens GPT-5.6 Sol generates.
- Separate standard, cached and long-context usage.
- Apply the matching rate to each category rather than using one blended percentage.
API billing is separate from ChatGPT subscriptions
The lower GPT-5.6 Sol figures are API rates for developer usage. They are not monthly prices for ChatGPT Pro, Plus or Business, and an API price change should not be treated as an automatic change to subscription usage limits.
That distinction is easy to miss because the same model family can appear across developer tools and consumer-facing products. The billing systems and units are different: the API charges by token categories, while a subscription is sold as a plan with its own access and allowance rules.
The same caution applies to Codex and ChatGPT Work credits. Their relationship to API billing should not be reduced to a simple dollar-for-token conversion when estimating subscription usage.
How long are the promotional rates listed to last?
OpenAI lists GPT-5.6 Sol’s promotional pricing as available at least through November 21, 2026. That date belongs to the promotional schedule, not to a permanent price guarantee.
For a production application, the sensible approach is to keep the rate category in your cost model and revisit the schedule before committing to a long-term budget. The bottom line is clear: GPT-5.6 Sol is cheaper on the listed standard short-context API rates, but the real saving depends on what your application sends, what it generates and whether its requests use caching or long context.
