GPT-6 Sol cost model

Updated September 23, 2026 · first published September 23, 2026

OpenAI released GPT-6 Sol on 22 September 2026. OpenAI positions it as the everyday model for complex work, coding included, and says it makes about half as many mistakes as GPT-5.6 Sol. For a cost owner, the headline is simpler: it costs half as much per token as the model it replaces.

GPT-6 Sol prices

TierInputCached inputOutput
Standard$2.00$0.20$10.00
Batch or flex$1.00$0.10$5.00
Fast mode$4.00$0.40$20.00
Long context (over 272K input)$4.00$0.40$15.00

Per 1M tokens, from OpenAI's pricing page. A cache write costs $2.50. The model has a 1,050,000-token context window, 128,000 max output tokens and an April 2026 knowledge cutoff.

Against GPT-5.6 Sol and Claude Opus 5.5

ModelInputCachedOutput
GPT-6 Sol$2.00$0.20$10.00
GPT-5.6 Sol$4.00$0.40$20.00
Claude Opus 5.5$4.00$0.20$20.00

GPT-5.6 Sol's $4/$20 is itself a promotional price that OpenAI says runs at least through 21 November 2026, so the gap may widen after that. Against Opus 5.5, Sol is half the price on fresh input and output, but cached input costs the same $0.20. For an agent that re-reads a large cached context and writes little, the two models cost closer to each other than the headline prices suggest.

Cost per task at each effort level

Per-token price is only half the bill. The other half is how many tokens the model spends thinking. Artificial Analysis ran GPT-6 Sol through its Intelligence Index (v4.3.2) at every effort setting.

EffortIndex scoreCost per task
none28.1$0.33
low33.9$0.13
medium (default)39.8$0.25
high42.8$0.37
xhigh44.1$0.53
max47.5$1.06

Two things stand out. First, none is not the cheap setting: it scored lower than low and cost 2.5× more per task on this index. Measure on your own traffic before assuming that turning reasoning off saves money. Second, going from medium to max adds 7.7 points at 4.2× the cost. Every setting from low to max sits on the price-performance frontier, so none is wasted, but the last step (xhigh to max) doubles the cost for 3.4 points.

For comparison, Claude Opus 5.5 at medium effort scores 51.2 at $1.34 per task. That is higher than any Sol setting. If your workload needs that score, Sol cannot reach it at any price. If it does not, Sol at high effort (42.8 for $0.37) costs about a quarter as much. Full numbers across models are in real cost per task.

What an agent step costs

A step with 200K input tokens, 90% served from cache, and 20K output tokens:

Output tokens are 72% of the Sol figure ($0.20 of $0.276). Capping output length and choosing a lower effort setting move this bill more than further cache tuning.

Watch the 272K line

Above 272K input tokens the whole request is billed at 2× input and 1.5× output. The same step at 270K input costs $0.303. At 300K it costs $0.528, 74% more for 11% more context. A long-running Sol agent that never compacts its history will cross that line on its own. Alert on it.

When to move to GPT-6 Sol

Sources: OpenAI GPT-6 Sol model page, TechCrunch, MacRumors, Artificial Analysis. Prices read 23 September 2026.

Related


Want this applied to your own LLM spend? FinOps LLM runs a free audit of your AI costs and shows where the savings are. Book free audit →

Back to research