OpenAI
gpt-4.1-2025-04-14
Made by OpenAI. Released 2025-04-14. Accessibility not recorded. Cheapest standard rate we hold: $12 per million output tokens, no context split.
What it costs
| Tier | Kind | In | Out | Cached in | From |
|---|---|---|---|---|---|
| batch | text | $1.50 | $6 | $0.50 | seller |
| standard | text | $3 | $12 | $0.75 | seller |
per million tokens, US dollars. * marks a rate from a single registry or two that disagree — the best available number rather than a confirmed one. Every other row is confirmed against a seller or a licensed index.
What it scores, by house
| Benchmark | House | Score | Evaluated | Scaffold |
|---|---|---|---|---|
| chess puzzles | Epoch AI | 0.1±0.02 | 2026-08-07 | — |
| frontiermath | Epoch AI | 0.1±0.01 | 2025-04-14 | — |
| frontiermath tier 4 | Epoch AI | 0.0±0.00 | 2025-07-01 | — |
| frontiermath tiers 1 3 v2 | Epoch AI | 0.1±0.01 | 2026-08-27 | — |
| gpqa diamond | Epoch AI | 0.7±0.03 | 2025-04-14 | — |
| math level 5near its ceiling | Epoch AI | 0.8±0.01 | 2025-04-14 | — |
| otis mock aime 2024 2025near its ceiling | Epoch AI | 0.4±0.06 | 2025-04-14 | — |
| simpleqa verified | Epoch AI | 0.3±0.01 | 2026-08-31 | — |
| swe bench verified | Epoch AI | 0.5±0.02 | 2026-02-08 | — |
| aider polyglot | collected by Epoch AI | 52.4 | — | — |
| ale bench | collected by Epoch AI | 558 | — | — |
| arc aginear its ceiling | collected by Epoch AI | 0.1 | — | — |
| arc agi 2 | collected by Epoch AI | 0.0 | — | — |
| cad eval | collected by Epoch AI | 0.4 | — | — |
| dtbenchnear its ceiling | collected by Epoch AI | 0.7 | — | — |
| enigma eval | collected by Epoch AI | 0.0 | — | — |
| fictionlivebench | collected by Epoch AI | 0.6 | — | — |
| forecastbench | collected by Epoch AI | 61.5 | — | — |
| geobench | collected by Epoch AI | 3791 | — | — |
| hle | collected by Epoch AI | 0.1 | — | — |
| lmca | collected by Epoch AI | 25.6 | — | — |
| simplebench | collected by Epoch AI | 0.3 | — | — |
| weirdml | collected by Epoch AI | 0.4 | — | — |
These scores are not comparable between houses and are not added up. A score is dated by when it was evaluated, not when the model was released, and a score without its scaffold is not a measurement — which is why both are printed.
Every rate we hold · Why there is no single best model · Everything from OpenAI