Code World Model (CWM)
Code World Model (CWM) places in the top three on CRUXEval.
28
Benchmarks with a score
0
Ranked first, out of five or more models
1
In the top three
Meta Sep 2025 Code World Model (CWM)
Open weights Mid-size
No Epoch Capabilities Index score yet
- LiveCodeBench v5 68.6% #13 /44
- Virology Capabilities Test (VCT) 23.8% #28 /28
- BioLP-bench 17.7% #9 /9
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
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GSM8K | 93.3% | #22 of 87 | Model card | |
| MATH-500 | 96.6% | #26 of 67 | Model card | |
| AIME 2024 | 76% | #51 of 87 | Model card | |
| AIME 2025 | 68.2% | #115 of 143 | Model card | |
| GPQA Diamond | 40.6% | #255 of 284 | Model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CRUXEval | 94.3% | #2 of 9 | Model card | |
| LiveCodeBench v5 | 68.6% | #13 of 44 | Model card | |
| MBPP | 73.4% | #32 of 66 | Model card | |
| SWE-bench Verified | 65.8% | #96 of 150 | Model card | |
| HumanEval+ | 75% | #21 of 27 | Model card | |
| Aider Polyglot | 35.1% | #24 of 29 | Model card | |
| LiveCodeBench v6 | 63.5% | #63 of 76 | Model card | |
| HaltEval-prelim (Meta) | 94% | #1 of 1 | Model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Terminal-Bench | 26.3% | #22 of 22 | Model card |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MASK | 44.8% | #17 of 21 | Model card | |
| Cybench | 25% | #16 of 17 | Model card | |
| Virology Capabilities Test (VCT) | 23.8% | #28 of 28 | Model card | |
| BioLP-bench | 17.7% | #9 of 9 | Model card | |
| Human Pathogen Capabilities Test | 31.2% | #8 of 8 | Model card | |
| Molecular Biology Capabilities Test | 32.7% | #7 of 7 | Model card | |
| WMDP Biology | 78.1% | #7 of 7 | Model card | |
| WMDP Chemistry | 64.6% | #5 of 5 | Model card | |
| Hack the Box Machines (Meta) | 41% | #1 of 1 | Model card | |
| Meta BioProtocol Proxy | 43.6% | #1 of 1 | Model card | |
| Native Code Exploitation (Meta) | 8.30% | #1 of 1 | Model card | |
| WMDP Cyber | 63.6% | #1 of 1 | Model card |
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.
Its system cards
Every result in the longer documents the lab published with it. 12 of them aren't in the launch post.
Capabilities
- SWE-bench Verified 65.8
- Aider Polyglot 35.1
- Terminal-Bench 26.3
- CRUXEval 94.3
- HaltEval-prelim (Meta) 94
- BigO(Bench)
- LiveCodeBench v5 68.6
- LiveCodeBench v6 63.5
- MATH-500 96.6
- AIME 2024 76
- AIME 2025 68.2
- MMLU 77.7
- MMLU-Pro 60.2
- GPQA Diamond 40.6
- GSM8K 93.3
- HumanEval+ 75
- MBPP 73.4
- RULER
Code World Model Preparedness Report 8 Oct 2025
Alignment
- MASK 44.8