Model economics
Price-per-token tables tell you what a model charges. They can't tell you what it ends up costing, because that depends on what people reach for it to do. These figures are measured from real sessions submitted by developers who track their own usage.
| Model | Per developer | Share of spend |
|---|---|---|
| gpt-5.6-sol | $10,996 | 20% |
| claude-opus-5 | $5,868 | 10% |
| claude-opus-4-8 | $5,387 | 20% |
| gpt-5.5 | $5,275 | 15% |
| claude-fable-5 | $5,093 | 17% |
| claude-opus-4-6 | $1,721 | 7% |
| claude-opus-4-7 | $1,600 | 5% |
| gpt-5.4 | $795 | 2% |
| claude-sonnet-5 | $565 | 1% |
| gpt-5.3-codex | $295 | 0.5% |
| gpt-5.6-terra | $189 | 0.3% |
| claude-sonnet-4-6 | $149 | 0.7% |
| claude-opus-4-1 | $137 | 0.4% |
| claude-sonnet-4-5 | $54 | 0.3% |
| claude-haiku-4-5 | $51 | 0.4% |
Models with fewer than 20 observed developers are omitted — too few to say anything honest about. Method and a free JSON endpoint on the data page.