o4-mini-high Pricing Explained: Historical Rates and API Migration
This is a historical pricing and migration guide, checked October 23, 2026. "o4-mini-high" was a ChatGPT option, not a separately priced API model. OpenAI's deprecation table lists October 23, 2026 as the shutdown date for o4-mini and its o4-mini-2025-04-16 snapshot. This update did not call the API to confirm the endpoint's actual status. The notice gives a date without a shutdown time or timezone.
What "high" meant
OpenAI's April 2025 announcement introduced o4-mini and the o4-mini-high option in ChatGPT. The high option allowed more reasoning effort. The API model was o4-mini; o4-mini-high was not an additional model identifier to put in the model field.
Reasoning effort is a setting on supported models. The current reasoning guide uses reasoning.effort in Responses and reasoning_effort in Chat Completions. Available settings depend on the model. A high setting may consume more reasoning tokens and take longer, but it does not establish a separate universal "high" price.
Historical o4-mini API prices
The retained, deprecated o4-mini model card lists these standard text-token rates in US dollars per million tokens:
| Token category | Historical rate |
|---|---|
| Input | $1.10 |
| Cached input | $0.275 |
| Output | $4.40 |
These are historical o4-mini rates, not a quote for its replacement or proof that the old endpoint still accepts requests. The cached rate only applies to input that qualifies as cached; it is not a blanket discount on every repeated prompt. Other service tiers and applicable tools can change a bill.
Reasoning tokens are billed as output even though they are not shown as answer text. For a purely arithmetic example, 1,000 uncached input tokens plus 2,000 output tokens would cost $0.0099 at those historical standard rates: $0.0011 input plus $0.0088 output. If 1,500 of those output tokens were reasoning, only 500 might be visible, while all 2,000 remain part of the output charge. This is an illustration, not a measured API request, and excludes extra tools and taxes.
ChatGPT and API bills are separate
A ChatGPT subscription does not purchase that many API tokens. OpenAI maintains separate billing systems for ChatGPT and the API platform. For an API charge, inspect the API organization's billing and usage rather than looking for an o4-mini-high line in a ChatGPT subscription invoice. Model choice, token consumption, tools and storage can each matter.
The original announcement now notes that o3 and o4-mini are no longer available in ChatGPT. Its old plan eligibility and model-picker screenshots should not be used as current subscription advice.
Migrate an existing integration
The current shutdown table recommends gpt-5.6-terra. The older o4-mini model card names GPT-5 Mini as its successor; for this retirement, use the dated deprecation schedule as the starting point.
- Find configuration entries using
o4-mini, its snapshot, or a third-party "high" label, and check which provider actually handles them. - Read the replacement's model documentation and current API pricing. Do not carry over o4-mini rates or assume identical effort values and tool support.
- Compare representative prompts, structured outputs, tool calls and latency before switching production traffic.
- After migration, inspect returned usage, including reasoning tokens, and verify the actual charges in the API dashboard.
Changing the model identifier alone does not validate behaviour, cost or availability. This page preserves the old search topic while directing new integrations to current documentation.