OpenAI cuts API prices in half with new Flex mode

Running heavy background AI tasks has always felt a bit unfair. If you are doing offline model evaluations or overnight data enrichment, nobody is sitting there waiting for an instant reply. Yet, you still had to pay the exact same premium price as someone powering a live chatbot.

OpenAI is finally fixing that math. On April 17, the company rolled out Flex processing, a new API option that cuts prices right down the middle in exchange for slower response times and occasional resource downtime.

Half the price, if you can wait

Flex processing is launching in beta for OpenAI's newly released reasoning models, o3 and o4-mini.

OpenAI says the tier is built specifically for low-priority, non-production tasks. Think model evaluations, synthetic data enrichment, and asynchronous background pipelines where a delayed response won't break anything.

The savings are substantial. For the flagship o3, Flex drops the price from the standard $10 per million input tokens and $40 per million output tokens down to $5 and $20.

For the budget-friendly o4-mini, the price falls to $0.55 per million input tokens and $2.20 per million output tokens, halved from its regular $1.10 and $4.40 rates.

The fine print: ID checks required

There is a catch if you want to take advantage of these lower rates on the flagship tier. In an email sent to customers, OpenAI noted that developers in tiers 1 through 3 will need to complete a newly introduced ID verification process to access o3.

These developer tiers are based on cumulative platform spend. On top of model access itself, reasoning summaries and streaming API support for o3 and select other models are now also locked behind that verification wall.

OpenAI previously stated that requiring government ID is meant to keep malicious actors from abusing its safety policies, but it adds clear friction for smaller builders.

Why it matters

We think this move is a clear reaction to mounting price pressure across the entire frontier AI space.

Just this Thursday, Google unleashed Gemini 2.5 Flash, a reasoning model designed to match or beat DeepSeek's R1 while keeping input token costs extremely aggressive. With rivals offering high reasoning performance at bargain basement rates, OpenAI had to find a way to make its expensive frontier models accessible without publicly devaluing its standard tier.

Borrowing a page from traditional cloud computing's spot instances is a smart play. It lets OpenAI monetize off-peak data center capacity while keeping developer workloads locked inside its ecosystem.

If you run heavy asynchronous pipelines, evals, or bulk processing, switching over to Flex is a no-brainer that immediately slashes your invoice by 50%. Just don't let it anywhere near customer-facing real-time applications, where intermittent unavailability will instantly backfire.