Gemini 3 Flash Preview
par Google DeepMind · Sorti le 2025-12-17
50.5
score moyen
$0.50/1M
Prix d'entrée
$3.00/1M
Prix de sortie
1.0M tokens (~524 books)
Fenêtre de contexte
multimodal
Type
Tested on 15 benchmarks with 50.5% average. Top scores: OTIS Mock AIME 2024-2025 (92.8%), GeoBench (88.0%), GPQA diamond (77.6%).
Scores de benchmark
| Benchmark | Catégorie | Score | Bar |
|---|---|---|---|
| OTIS Mock AIME 2024-2025 | math | 92.8 | |
| GeoBench | knowledge | 88.0 | |
| GPQA diamond | knowledge | 77.6 | |
| SimpleQA Verified | knowledge | 67.4 | |
| Terminal Bench | coding | 64.3 | |
| WeirdML | coding | 61.6 | |
| VPCT | knowledge | 58.9 | |
| SimpleBench | reasoning | 53.3 | |
| Balrog | knowledge | 48.1 | |
| Chess Puzzles | knowledge | 38.0 | |
| FrontierMath-2025-02-28-Private | math | 35.6 | |
| ARC-AGI-2 | reasoning | 33.6 | |
| APEX-Agents | agentic | 24.0 | |
| GSO-Bench | coding | 9.8 | |
| FrontierMath-Tier-4-2025-07-01-Private | math | 4.2 |