Model
DeepSeek V3.2
DeepSeek · open weights
Knowledge & Reasoning
Knowledge & Reasoning| Benchmark | Score | Metric | Harness | Source | Confidence | Link |
|---|---|---|---|---|---|---|
| CL-bench | 13.2 | overall | N/A | Independent | low | epoch.ai |
| CL-bench Life | 9.5 | overall | N/A | Independent | low | epoch.ai |
| Colombian Legal Reliability Benchmark | 72.8 | accuracy | N/A | Independent | medium | arxiv.org |
| CritPt | 2.9 | accuracy | N/A | Independent | low | epoch.ai |
| CroissantMiner | 0.6 | accuracy | single-pass | Independent | medium | arxiv.org |
| GPQA Diamond | 79.9 | accuracy | N/A | Self-reported | low | artificialanalysis.ai |
| LMArena | 1423.0 | elo | N/A | Self-reported | low | lmarena.ai |
| MMLU-Pro | 85.0 | accuracy | N/A | Self-reported | low | artificialanalysis.ai |
Math
Math| Benchmark | Score | Metric | Harness | Source | Confidence | Link |
|---|---|---|---|---|---|---|
| AIME 2025 | 89.3 | accuracy | N/A | Self-reported | low | kaggle.com |
Coding
Coding| Benchmark | Score | Metric | Harness | Source | Confidence | Link |
|---|---|---|---|---|---|---|
| Aider Polyglot | 74.2 | percent_correct | N/A | Independent | medium | aider.chat |
| LiveCodeBench | 74.1 | pass@1 | N/A | Self-reported | low | livecodebench.github.io |
| QTestBench | 0.4 | test_quality | N/A | Independent | medium | arxiv.org |
| SciCode | 38.9 | score | N/A | Independent | medium | artificialanalysis.ai |
| SWE-bench Verified | 67.8 | resolved_rate | N/A | Self-reported | low | swebench.com |
| Terminal-Bench 2.0 | 39.6 | accuracy | N/A | Self-reported | low | tbench.ai |
| VeriCodeBench | 17.0 | success_rate | N/A | Independent | medium | arxiv.org |
Agentic / Tool Use
Agentic / Tool Use| Benchmark | Score | Metric | Harness | Source | Confidence | Link |
|---|---|---|---|---|---|---|
| FinSkillBench | 0.5 | score | N/A | Independent | medium | arxiv.org |
| GeoNatureAgent (GNA) | 56.3 | accuracy | N/A | Independent | medium | arxiv.org |
| KaliBench | 80.2 | total_score_(%)_avg. | N/A | Independent | medium | arxiv.org |
| LiveMACEBench | 0.6 | ras | N/A | Independent | medium | arxiv.org |
| MASBench | 22.5 | score | N/A | Independent | medium | arxiv.org |
| TheAgentCompany | 42.9 | resolved_rate | N/A | Independent | medium | the-agent-company.com |
| XiangqiBench | 0.8 | pass@3 | N/A | Independent | medium | arxiv.org |
Safety & Alignment
Safety & Alignment| Benchmark | Score | Metric | Harness | Source | Confidence | Link |
|---|---|---|---|---|---|---|
| Jailbreak Guardrails Benchmark | 40.3 | bypass_rate | N/A | Independent | medium | arxiv.org |
| Live-Harness-Bench | 13.8 | success_rate | live-harness-bench | Independent | medium | arxiv.org |
Reproduced by the community
Loading community runs…