GPT-5.6 Sol drops to $4 and $20, and overtakes Claude Opus 5 on cost

Price changes are the part of a model's specification that moves fastest and gets checked least, so here is the whole of it. On August 21 OpenAI reduced the API price of GPT-5.6 Sol by 20% on input and 33% on output. Everything else about the model is unchanged. The reduction is explicitly a promotion with an end date attached.

The new numbers

Input falls from $5 to $4 per million tokens, cached input from $0.50 to $0.40, and output from $30 to $20.

The same promotion covers the two smaller models in the family. Terra goes from $2.50 to $2.00 on input and from $15 to $12 on output. Luna goes from $1.00 to $0.20 on input and from $6.00 to $1.20 on output, the steepest cut of the three and the one to read twice if you run high volume.

What it changes in practice

Price a single task the way a bill actually accumulates — 4,000 tokens in, 800 out, a thousand times — and Sol goes from $44.00 to $32.00, a drop of 27%. Claude Opus 5, unchanged at $5 and $25, costs $40.00 for the same thousand calls.

That is the finding worth carrying. Before August 21 the choice between the two frontier models was between $44 and $40, and Sol was the more expensive one. Today it is $32 against $40, and Sol is the cheaper one by a fifth. Nothing about either model's behavior moved; the ranking flipped on a price change alone.

The input columns crossed too. Sol and Opus 5 charged the same $5 per million on input, which made input a wash and pushed every comparison onto the output column. Now Sol is a dollar cheaper on input and five dollars cheaper on output.

The date attached to it

OpenAI calls this promotional pricing and says it is available at least through November 21, 2026, describing it internally as a three-month promotion running from August 21. The phrase to notice is at least: the floor is stated, the ceiling is not.

For anything you plan to still be running in December, that makes this a rate to budget twice — once at $4 and $20, and once at whatever follows. It is the same discipline the Gemini Flash rows already demand, and the reason a price without a date is not a specification.

Where it does not apply

The reduction covers the API, including Fast mode and the Batch and Flex processing tiers, and is rolling out across eligible plans for ChatGPT Work and Codex credits. OpenAI states that Pro, Plus and Business subscription usage remains unchanged, so a person paying monthly for ChatGPT sees nothing.

This is a change for whoever is billed by the token.