Llama 4 Maverick
オープンソース開発元 Meta · リリース日 2025-04-05
28.0
平均スコア
$0.15/1M
入力料金
$0.60/1M
出力料金
1.0M tokens (~524 books)
コンテキストウィンドウ
multimodal
タイプ
Tested on 17 benchmarks with 28.0% average. Top scores: MATH level 5 (73.0%), Lech Mazur Writing (63.7%), GPQA diamond (56.0%).
ベンチマークスコア
| ベンチマーク | カテゴリ | スコア | Bar |
|---|---|---|---|
| MATH level 5 | math | 73.0 | |
| Lech Mazur Writing | knowledge | 63.7 | |
| GPQA diamond | knowledge | 56.0 | |
| GeoBench | knowledge | 52.0 | |
| Fiction.LiveBench | knowledge | 46.2 | |
| WeirdML | coding | 24.5 | |
| SWE-Bench Verified (Bash Only) | coding | 21.0 | |
| OTIS Mock AIME 2024-2025 | math | 20.5 | |
| Artificial Analysis — Quality Index | speed | 18.4 | |
| Aider polyglot | coding | 15.6 | |
| Artificial Analysis — Coding Index | speed | 15.6 | |
| SimpleBench | reasoning | 13.2 | |
| Artificial Analysis — Agentic Index | speed | 7.2 | |
| ARC-AGI | reasoning | 4.4 | |
| HLE | knowledge | 0.9 | |
| FrontierMath-2025-02-28-Private | math | 0.7 | |
| ARC-AGI-2 | reasoning | 0.1 |