OpenAI has continued its push to make access to its flagship GPT-5.6 lineup more affordable. If at the end of July the company had already revised the pricing for two models in the series — Luna and Terra — while also expanding their usage limits, then now, according to Reuters, similar changes have been applied to the third and most performant of the trio, the Sol model, which is aimed at software developers. For the next three months, the cost of access to it will be reduced by more than 20%.
What exactly got cheaper
It is important to distinguish between two pricing tiers. The cost of consumer and enterprise subscriptions — the Pro, Plus, and Business plans — has remained unchanged. Only access to the model via the API has been adjusted, and the new terms will be available to ChatGPT Work and Codex customers. The specific figures look like this: for every million input tokens, users of GPT-5.6 Sol now pay $4, while output tokens under standard use with a short context cost $20 per million. Before the revision, these rates were $5 and $30, respectively.
Competitive context
To gauge how aggressive OpenAI's move is, it is useful to compare its rates with those of its main rival. Anthropic offers access to its flagship Claude Fable 5 model at $10 per million input tokens and $50 per million output tokens, while working with Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens. Thus, even after the cut, Sol remains noticeably cheaper than Anthropic's top model, but is comparable to its more affordable Opus 5 on input tokens and cheaper on output tokens.
Reasons: competition and IPO
AI developers are being forced to adjust their prices by growing competition in the API market and the need to demonstrate positive revenue trends in light of upcoming public offerings. At the same time, the timelines of the two leaders diverge: Anthropic, according to available reports, is determined to go public as early as this fall, whereas OpenAI is prepared to wait until next year. It is precisely this difference in IPO timing that may explain OpenAI's tougher pricing policy in the coming months.
Contradictory data
The phrasing "a reduction of more than 20%" contains a slight inconsistency with the actual figures. For input tokens, the rate dropped exactly from $5 to $4, i.e. by 20%, and not "more than 20%". The 20% threshold is exceeded only thanks to output tokens, where the reduction amounted to a third — from $30 to $20. In other words, the "more than 20%" figure cited in the news is valid only as an average or as a characteristic of output traffic, whereas for input tokens the reduction is exactly 20%. Readers should keep this difference in mind when calculating their own cost of using the API.
