Epoch
Gemini 3 Flash Preview
Made by Google DeepMind. Released 2025-12-17. Accessibility not recorded. Cheapest standard rate we hold: $3 per million output tokens, no context split.
What it costs
| Tier | Kind | In | Out | Cached in | From |
|---|---|---|---|---|---|
| batch | text | $0.25 | $1.50 | — | seller |
| flex | text | $0.25 | $1.50 | — | seller |
| priority | text | $0.90 | $5.40 | — | seller |
| standard | text | $0.50 | $3 | — | 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 | 0.4±0.05 | 2026-08-06 | high |
| chess puzzles | Epoch AI | 0.4±0.05 | 2025-12-17 | — |
| frontiermath | Epoch AI | 0.4±0.03 | 2025-12-17 | — |
| frontiermath tier 4 | Epoch AI | 0.0±0.03 | 2025-12-17 | — |
| frontiermath tier 4 v2 | Epoch AI | 0.2±0.06 | 2026-06-11 | — |
| frontiermath tiers 1 3 v2 | Epoch AI | 0.5±0.03 | 2026-06-11 | — |
| gpqa diamond | Epoch AI | 0.9±0.02 | 2026-08-06 | high |
| gpqa diamond | Epoch AI | 0.8±0.02 | 2025-12-17 | — |
| mystery game puzzles | Epoch AI | 0.3±0.04 | 2026-08-05 | minimal |
| mystery game puzzles | Epoch AI | 0.3±0.04 | 2026-08-05 | low |
| mystery game puzzles | Epoch AI | 0.2±0.04 | 2026-08-05 | high |
| otis mock aime 2024 2025near its ceiling | Epoch AI | 1.0±0.03 | 2026-08-06 | high |
| otis mock aime 2024 2025near its ceiling | Epoch AI | 0.9±0.04 | 2025-12-17 | — |
| simpleqa verified | Epoch AI | 0.7±0.01 | 2026-08-27 | high |
| swe bench verified | Epoch AI | 0.8±0.02 | 2026-02-18 | — |
| ale bench | collected by Epoch AI | 1367 | — | — |
| apex agents | collected by Epoch AI | 0.4 | — | — |
| arc aginear its ceiling | collected by Epoch AI | 0.8 | — | — |
| arc agi 2 | collected by Epoch AI | 0.3 | — | — |
| balrog | collected by Epoch AI | 0.5 | — | — |
| blueprint bench 2 | collected by Epoch AI | 0.0 | — | — |
| deepresearchbench | collected by Epoch AI | 0.5 | — | minimal |
| deepresearchbench | collected by Epoch AI | 0.5 | — | high |
| deepresearchbench | collected by Epoch AI | 0.5 | — | low |
| dtbenchnear its ceiling | collected by Epoch AI | 0.9 | — | high |
| forecastbench | collected by Epoch AI | 58.5 | — | — |
| gdp pdf | collected by Epoch AI | 0.1 | — | — |
| gdp pdf | collected by Epoch AI | 0.1 | — | high |
| geobench | collected by Epoch AI | 4333 | — | — |
| gso | collected by Epoch AI | 0.1 | — | OpenHands |
| lmca | collected by Epoch AI | 43.1 | — | high |
| proofbench | collected by Epoch AI | 0.1 | — | — |
| simplebench | collected by Epoch AI | 0.6 | — | — |
| terminalbench | collected by Epoch AI | 0.6 | — | Junie CLI |
| terminalbench | collected by Epoch AI | 0.5 | — | Terminus 2 |
| terminalbench | collected by Epoch AI | 0.5 | — | Gemini CLI |
| vending bench 2 | collected by Epoch AI | 3635 | — | — |
| vpct | collected by Epoch AI | 0.7 | — | — |
| webdev arena | collected by Epoch AI | 1383 | — | — |
| weirdml | collected by Epoch AI | 0.6 | — | — |
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 Epoch