OpenAI cuts GPT-5.6 Sol API pricing by over 20% for three months

GPT-5.6 Sol is a reasoning-focused model OpenAI rolled out to Plus and Pro users on August 8, adding a slider to adjust reasoning depth alongside improvements to answer conciseness and factual accuracy. In the same rollout, free and Go tier users were defaulted to a separate model, GPT-5.6 Luna. A startup case study published on August 17 then showed real estate startup HipaAI cutting costs by a factor of 18 using GPT-5.6 Luna while maintaining 98% document extraction accuracy, while GPT-5.6 Sol raised its ARC-AGI-3 benchmark score from 13.3% to 38.3% while cutting output tokens by roughly a factor of six. The usage dashboard covered in OpenAI unveils per-API-key spend tracking and limits also launched the same day, and this price cut ties into that same trend.

How far does the cut extend

A follow-up tweet from the OpenAI Developers account laid out the reasoning behind the cut in more detail. It attributed the change to being able to run the model more efficiently, meaning purchased credits will now last longer under Codex's token-based plans. Usage included in subscription plans, however, remains unchanged. In the same thread, OpenAI's account added, to the effect that, "This applies to the API immediately, and rolls out to ChatGPT Work and Codex credits by plan. There's no change to Pro, Plus, or Business subscription usage."

Category	Change
API pricing	Cut by more than 20% immediately, in effect for three months
Codex credits (token-based plans)	Same spend now covers more usage
ChatGPT Work / Codex plans	Rolling out by plan
Pro / Plus / Business subscription usage	No change

Who feels the impact

Developers and businesses calling GPT-5.6 Sol directly through the API will see the cut reflected immediately in their bills. Users on Codex's token-based plans will also be able to do more work on the same budget. Users on Pro, Plus, or Business flat-rate ChatGPT subscriptions, on the other hand, aren't affected — the included usage OpenAI provides them stays the same, as explicitly stated.

Editor's view

The fact that OpenAI announced this price cut through two tweets rather than a pricing page tells you something about where the company's focus is right now. This is a move out of the developer-facing API and agent-infrastructure race, not the consumer ChatGPT business. OpenAI is backing up its claim that it can now run the model more cheaply with an actual price cut, and choosing to pass those savings on to developers.

Just last month, when the reasoning-depth slider was added to GPT-5.6 Sol, the model read as one that thought more carefully but cost more to run. But the HipaAI case showed GPT-5.6 Luna cutting costs by a factor of 18 while holding accuracy steady, and Sol nearly tripling its ARC-AGI-3 score while cutting output tokens by a factor of six. This price cut is the first instance of that efficiency gain actually showing up in the price sheet.

Development teams in Korea running Codex on token-based plans will immediately notice they can do more work on the same budget. But two conditions are worth keeping in mind: the cut is only in effect for three months, and it doesn't apply to Pro, Plus, or Business flat-rate subscriptions. Teams calling GPT-5.6 Sol directly through the API can use the per-API-key spend tracking dashboard OpenAI released the same day to directly compare costs before and after the cut.

Having committed to a fixed three-month window, OpenAI will likely roll out further pricing moves within that period aimed at pulling more enterprise spend away from Anthropic.