NEWS

OpenAI cuts GPT-5.6 Sol API price by more than 20%

OpenAI's frontier model drops to $4 per million input tokens and $20 on output, in a three-month window that directly affects the cost of AI projects in Brazil.

OpenAI cuts GPT-5.6 Sol API price by more than 20%
Image: Redação iMasters

OpenAI announced on Friday that it will cut the price of its frontier model GPT-5.6 Sol for developers by more than 20%, valid for three months. The cut applies to the company's API and is being distributed through eligible plans, including credits for the agentic AI product ChatGPT Work and the coding tool Codex, according to Reuters, published by Economic Times.

The move comes amid growing competitive pressure from Anthropic and Chinese models, according to the same source.

The new numbers

Under OpenAI's pricing table, GPT-5.6 Sol now costs:

  • $4 per 1 million input tokens (down from $5)
  • $20 per 1 million output tokens (down from $30)

The figures refer to standard short-context use. In practice, the output token, usually the most expensive line item in any application's bill, dropped by a third.

It's important to note what did not change: prices for the Pro, Plus, and Business subscriptions remain unchanged. The cut is exclusive to those who consume via API, that is, the audience that builds software on top of the model.

Not the first cut of the quarter

The Sol adjustment comes after OpenAI changed pricing for its smaller models at the end of the previous month. According to Economic Times, the company cut:

  • 20% on GPT-5.6 Terra (mid-tier model)
  • 80% on Luna (lowest-cost model)

In other words, the reduction on the frontier model completes a pricing sweep that was already underway across the entire lineup, from the cheapest to the most expensive.

How it compares to Anthropic

The source draws a direct comparison with the competitor that has been troubling OpenAI the most. Under Anthropic's pricing table:

  • Claude Fable 5 (frontier): $10 per million input tokens and $50 on output
  • Claude Opus 5: $5 per million input tokens and $25 on output

With the cut, GPT-5.6 Sol at $4 / $20 comes in below even Opus 5, Anthropic's mid-tier model, and at less than half the price of the top-of-the-line Claude Fable 5. OpenAI's move is clear in the source: reposition its flagship model at an aggressive price point for a limited period.

What changes for AI builders in Brazil

For Brazilian startups and teams already running production on GPT-5.6 Sol, the effect on variable costs is immediate. An application that processes, for example, a lot of output tokens (verbose chatbots, text generation, agents that produce long responses) feels the drop from $30 to $20 more than the drop on the input side.

A simple calculation illustrates this: a service generating 500 million output tokens per month used to pay $15,000 on that front alone; at the new price, it drops to $10,000. That's $5,000 a month that, once converted and added to exchange-rate costs, weighs on the cash flow of any lean operation.

For Brazilian developers, it's also worth remembering two points the source doesn't cover, but which are part of the context of running a dollar-denominated API here:

  • Exchange rates and IOF (Brazil's tax on foreign-exchange transactions) factor into the real cost, so the nominal cut in dollars doesn't translate 1:1 into the bill in reais.
  • The benefit is tied to credits within eligible plans and specific products (ChatGPT Work and Codex), so those using the raw API should check the official pricing table to see if the reduced price applies to their case.

What remains open

The most sensitive point is the timeline. The reduction is valid for three months, according to Reuters, and the source doesn't detail what happens afterward: whether the price returns to its previous level, is extended, or becomes permanent. For those resizing architecture or renegotiating budgets based on this value, it's a data point that needs to be factored into planning with caution, since a reversal at the end of the window would push the output cost back from $20 to $30.

There's also no detail on long-context pricing, prompt caching, or higher usage tiers, which usually have their own rules. It's worth following OpenAI's official pricing table for the full picture before finalizing any ROI calculations.

Translated from the Brazilian Portuguese original · Read the original