Model
Claude Haiku 4.5
Anthropic · closed weights
Knowledge & Reasoning
Knowledge & Reasoning| Benchmark | Score | Metric | Harness | Source | Confidence | Link |
|---|---|---|---|---|---|---|
| Balrog | 31.2 | average_progress | N/A | Independent | medium | balrogai.com |
| CritPt | 0.0 | accuracy | N/A | Independent | low | epoch.ai |
| GPQA Diamond | 60.5 | accuracy | N/A | Independent | high | epoch.ai |
| MS-Exam-Gen | 85.6 | accuracy | N/A | Independent | medium | arxiv.org |
| SimpleQA Verified | 13.2 | accuracy | N/A | Independent | high | epoch.ai |
Coding
Coding| Benchmark | Score | Metric | Harness | Source | Confidence | Link |
|---|---|---|---|---|---|---|
| AptMQL-Bench | 64.7 | soft-ex | N/A | Independent | medium | arxiv.org |
| ARCS | 28.6 | ex_e2e | N/A | Independent | medium | arxiv.org |
| ArrivalBench | 52.2 | failure_rate | N/A | Independent | medium | arxiv.org |
| Hob-VL | 49.4 | accuracy | N/A | Independent | medium | arxiv.org |
| NP-Bench | 60.0 | clean-integration_success_rate | np-bench | Independent | medium | arxiv.org |
| ParanoiaEval | 17.7 | violation_rate | N/A | Independent | medium | arxiv.org |
| SciCode | 43.3 | score | N/A | Independent | medium | artificialanalysis.ai |
Agentic / Tool Use
Agentic / Tool Use| Benchmark | Score | Metric | Harness | Source | Confidence | Link |
|---|---|---|---|---|---|---|
| AgentHorizon | 52.2 | accuracy | claude code | Independent | medium | arxiv.org |
| Argo-Bench | 5.5 | score | N/A | Independent | medium | arxiv.org |
| DeepResearch Bench | 45.5 | average_score | N/A | Independent | medium | evals.futuresearch.ai |
| DrugTargetWorld | 12.9 | score | N/A | Independent | medium | arxiv.org |
| ExploitBench | 13.7 | mean_capability | N/A | Independent | low | epoch.ai |
| KlinikeBench | 2.7 | strict_pass@1 | N/A | Independent | medium | arxiv.org |
| MisVisAgentBench | 34.3 | pra | N/A | Independent | medium | arxiv.org |
| Terminal-Bench | 35.5 | accuracy | Goose | Independent | medium | tbench.ai |
| Terminal-Bench | 29.8 | accuracy | Mini-SWE-Agent | Independent | medium | tbench.ai |
| Terminal-Bench | 28.3 | accuracy | Terminus 2 | Independent | medium | tbench.ai |
| Terminal-Bench | 27.5 | accuracy | Claude Code | Independent | medium | tbench.ai |
| Terminal-Bench | 13.9 | accuracy | OpenHands | Independent | medium | tbench.ai |
Multimodal
MultimodalInstruction Following
Instruction FollowingSafety & Alignment
Safety & Alignment| Benchmark | Score | Metric | Harness | Source | Confidence | Link |
|---|---|---|---|---|---|---|
| DecepEval | 66.5 | deception_rate | N/A | Independent | medium | arxiv.org |
| MediateRec | 12.2 | resc. | N/A | Independent | medium | arxiv.org |
| Unified Misuse Monitoring Benchmark | 0.9 | auc | N/A | Independent | medium | arxiv.org |
Abstract Reasoning
Abstract ReasoningReproduced by the community
Loading community runs…