GPT-5 API pricing at a glance
OpenAI’s pricing documentation lists GPT-5 at $1.25 per 1 million input tokens, $0.125 per 1 million cached input tokens, and $10 per 1 million output tokens, in USD. These are the standard API rates in its pricing table—not ChatGPT subscription prices. (OpenAI pricing)
Availability: OpenAI’s deprecations notice says older GPT-5 and o3 model snapshots are scheduled for removal from the API on December 11, 2026. As of November 12, 2026, that date is upcoming, so treat the prices below as pre-shutdown reference rates—not a durable choice for a new production deployment. The notice refers to older snapshots; check the exact model ID’s shutdown_date using OpenAI’s Models API before relying on a particular ID. (OpenAI deprecations)
| GPT-5 token category | Standard price (USD per 1M tokens) |
|---|---|
| Input | $1.25 |
| Cached input | $0.125 |
| Output | $10.00 |
OpenAI’s model page also displays these prices beside a “Batch API price” label, while its pricing documentation lists them in the Standard pricing table. The rates above refer to the standard tier; don’t use them as a Batch estimate. (GPT-5 model page; OpenAI pricing)
Estimate GPT-5 API spend
Calculate uncached input, cached input, and output separately, then add them:
Estimated cost = (uncached input tokens ÷ 1,000,000 × $1.25) + (cached input tokens ÷ 1,000,000 × $0.125) + (output tokens ÷ 1,000,000 × $10)
For example, assume one request uses 100,000 uncached input tokens, no cached input, and 20,000 total output tokens as counted in API usage:
- Input: 100,000 ÷ 1,000,000 × $1.25 = $0.125
- Output: 20,000 ÷ 1,000,000 × $10 = $0.20
- Estimated total: $0.325 per request
The output count must include reasoning tokens: OpenAI says they count toward output usage and are billed as output tokens, even though they aren’t visible in the final answer. If 1,000 requests use the same token mix, the estimate is $325 at these standard rates. Cached input, when applicable, uses its separate rate. Tool-specific models can also have per-call fees, so include those where relevant. (OpenAI token guide; OpenAI pricing)
Compare GPT-5, mini, and nano rates
OpenAI’s standard pricing table lists these rates per 1 million tokens. OpenAI marks these model entries deprecated, and its deprecations notice schedules older GPT-5 snapshots for API removal on December 11, 2026. This table is therefore a comparison of published rates before the scheduled removal, not confirmation that these are suitable for a new deployment or will remain accessible after that date. The notice refers to older snapshots; check the exact ID’s shutdown_date in OpenAI’s Models API. (OpenAI pricing; OpenAI deprecations; GPT-5; GPT-5 mini; GPT-5 nano)
| Model | Input / 1M tokens | Cached input / 1M tokens | Output / 1M tokens | Status and documented positioning |
|---|---|---|---|---|
GPT-5 (gpt-5) |
$1.25 | $0.125 | $10.00 | Deprecated; previous model for coding, reasoning, and agentic tasks |
GPT-5 mini (gpt-5-mini) |
$0.25 | $0.025 | $2.00 | Deprecated; faster, more cost-efficient, suited to well-defined tasks and precise prompts |
GPT-5 nano (gpt-5-nano) |
$0.05 | $0.005 | $0.40 | Deprecated; lowest-cost variant, suited to summarization and classification |
Before the scheduled removal, mini has lower listed rates than GPT-5, while nano has the lowest listed rates and is positioned for simpler tasks such as summarization or classification. Those descriptions are model guidance, not a guarantee of quality on your workload. For a new production system, evaluate an available replacement rather than planning around these deprecated entries. (GPT-5 mini model page; GPT-5 nano model page; OpenAI deprecations)
Check your budget against actual usage
Use expected input, cached-input, and output token volumes to build an estimate, then multiply the per-request figure by your expected request count for a monthly projection. Reconcile that estimate with usage reported for actual API requests; visible answer length alone won’t capture reasoning-token usage. These figures are OpenAI API rates for the named models, not prices from an API gateway or reseller.