Llama 3.2 90B
Llama 3.2 90B ranks first on VL-RewardBench.
44
Benchmarks with a score
1
Ranked first, out of five or more models
2
In the top three
Meta Sep 2024 Llama 3.2 90B
Open weights
122.8
#178 of 256
Epoch Capabilities Index
- ZeroBench 0% #74 /76
- GPQA Diamond 41% #182 /229
- MicroVQA 42.4% #17 /19
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| EQ-Bench v2 | 82.0 | #16 of 85 | Official leaderboard | |
| MMBench-CN v1.1 | 75.2% | #77 of 178 | Official leaderboard | |
| AraGen | 50.5% | #20 of 42 | Official leaderboard | |
| Vectara Hallucination Leaderboard (HHEM) | 4.30% | #74 of 148 | Official leaderboard | |
| SEAL Precise Instruction Following | 82.1 | #12 of 18 | Official leaderboard | |
| SEAL Multilingual | 944 | #17 of 19 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MATH Level 5 | 39.4% | #63 of 94 | Epoch AI run | |
| GPQA Diamond | 41% | #182 of 229 | Epoch AI run | |
| U-MATH | 37.2% | #24 of 30 | Official leaderboard | |
| OTIS Mock AIME 2024-2025 | 2.64% | #190 of 212 | Epoch AI run |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| HumanEval-V | 5.90% | #16 of 24 | Official leaderboard | |
| SEAL Coding | 984 | #18 of 26 | Official leaderboard | |
| SArena | 0.76 | #17 of 17 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SnakeBench | 25 | #80 of 188 | Official leaderboard | |
| BALROG | 27.3% | #18 of 33 | Official leaderboard | |
| MCP-Bench | 0.49 | #14 of 15 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MedQA | 76.9% | #16 of 71 | Official leaderboard | |
| MortgageTax | 38.8% | #83 of 87 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| VL-RewardBench | 53.9% | #1 of 8 | Official leaderboard | |
| SEED-Bench-2-Plus | 68.2% | #28 of 136 | Official leaderboard | |
| CoreCognition | 60.6% | #26 of 119 | Official leaderboard | |
| MM-Vet | 64.1% | #43 of 164 | Official leaderboard | |
| DynaMath | 44% | #3 of 11 | Official leaderboard | |
| EmbodiedBench EB-Habitat | 40.3% | #5 of 17 | Official leaderboard | |
| MMBench-EN v1.1 | 79.5% | #58 of 178 | Official leaderboard | |
| MMBench | 80.4% | #52 of 155 | Official leaderboard | |
| CharXiv | 37.5% | #27 of 77 | Official leaderboard | |
| CCBench (MMBench Chinese Culture) | 54.1% | #54 of 152 | Official leaderboard | |
| MMMU | 60.3% | #54 of 136 | Official leaderboard | |
| MMStar | 55.3% | #80 of 173 | Official leaderboard | |
| EmbodiedBench EB-ALFRED | 32% | #8 of 17 | Official leaderboard | |
| MMVU | 47.1% | #14 of 28 | Official leaderboard | |
| VisualPuzzles | 34.1% | #16 of 29 | Official leaderboard | |
| EmbodiedBench EB-Manipulation | 14.9% | #10 of 17 | Official leaderboard | |
| MMMU-Pro | 39.5% | #35 of 57 | Official leaderboard | |
| EmbodiedBench EB-Navigation | 30% | #11 of 17 | Official leaderboard | |
| OmniAI OCR Benchmark | 54.2% | #3 of 4 | Official leaderboard | |
| GeoBench | 52% | #40 of 48 | Official leaderboard | |
| MicroVQA | 42.4% | #17 of 19 | Official leaderboard | |
| Vision Arena | 1,032 | #122 of 136 | Official leaderboard | |
| VISTA | 24.6% | #52 of 54 | Official leaderboard | |
| ZeroBench | 0% | #74 of 76 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| HELM Lite | 81.9 | #11 of 71 | Official leaderboard | |
| OpenVLM Leaderboard | 63.4 | #70 of 173 | Official leaderboard |
Meta's models over time
Meta's models in color, against every other model on the Epoch Capabilities Index, by release date.
Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)