Model profile
Claude 2
Anthropicdeveloper
2023-07-11release date
#3 / 267overall rank
6eval lineages
Evidence summary
Claude 2 has an estimated overall rank of #3; its 90% source-sensitivity interval is #1–#36. Its behavior-only rank is #4; company governance moves the combined estimate to #3. Published evidence spans 6 evals and 4 of 7 behavior components. Its strongest relative result is SALAD-Bench (base_representation_toxicity, #1 of 33); its weakest is SALAD-Bench (mcq_misinformation_harms, #23 of 33).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗claude-2
- OpenRouter ↗anthropic/claude-2
- Official model page ↗Exact model document · Anthropic · first party
Published eval results
Rank is within that sub-eval. Black marks the observed result; the grey dot marks the value implied by the global rank. Values stay on each source’s native scale.
| Eval / sub-eval | Rank | Value | Distribution | Better | Source |
|---|---|---|---|---|---|
| DecodingTrustmachine_ethics | #3 / 8 | 85.17 | ↑ higher | Source ↗official | |
| DecodingTruststereotype_bias | #1 / 8 | 100 | ↑ higher | Source ↗official | |
| DecodingTrusttoxicity | #1 / 8 | 92.11 | ↑ higher | Source ↗official | |
| Fake Alignment (FINE)multiple_choice_safe_decision_rate | #2 / 14 | 85.33 | ↑ higher | Source ↗official | |
| Fake Alignment (FINE)open_ended_safe_response_rate | #3 / 14 | 98.67 | ↑ higher | Source ↗official | |
| HarmBenchdr | #2 / 28 | 2 | ↓ lower | Source ↗official | |
| SALAD-Benchattack_enhanced_human_autonomy_integrity | #1 / 33 | 86.64 | ↑ higher | Source ↗official | |
| SALAD-Benchattack_enhanced_information_safety_harms | #1 / 33 | 93.49 | ↑ higher | Source ↗official | |
| SALAD-Benchattack_enhanced_malicious_use | #1 / 33 | 87.03 | ↑ higher | Source ↗official | |
| SALAD-Benchattack_enhanced_misinformation_harms | #1 / 33 | 91.61 | ↑ higher | Source ↗official | |
| SALAD-Benchattack_enhanced_representation_toxicity | #1 / 33 | 87.15 | ↑ higher | Source ↗official | |
| SALAD-Benchattack_enhanced_socioeconomic_harms | #1 / 33 | 88.31 | ↑ higher | Source ↗official | |
| SALAD-Benchbase_human_autonomy_integrity | #1 / 33 | 99.88 | ↑ higher | Source ↗official | |
| SALAD-Benchbase_information_safety_harms | #2 / 33 | 99.66 | ↑ higher | Source ↗official | |
| SALAD-Benchbase_malicious_use | #1 / 33 | 99.97 | ↑ higher | Source ↗official | |
| SALAD-Benchbase_misinformation_harms | #1 / 33 | 99.7 | ↑ higher | Source ↗official | |
| SALAD-Benchbase_representation_toxicity | #1 / 33 | 99.58 | ↑ higher | Source ↗official | |
| SALAD-Benchbase_socioeconomic_harms | #1 / 33 | 99.41 | ↑ higher | Source ↗official | |
| SALAD-Benchmcq_human_autonomy_integrity | #22 / 33 | 21.67 | ↑ higher | Source ↗official | |
| SALAD-Benchmcq_information_safety_harms | #20 / 33 | 28.61 | ↑ higher | Source ↗official | |
| SALAD-Benchmcq_malicious_use | #23 / 33 | 19.17 | ↑ higher | Source ↗official | |
| SALAD-Benchmcq_misinformation_harms | #23 / 33 | 20.71 | ↑ higher | Source ↗official | |
| SALAD-Benchmcq_representation_toxicity | #21 / 33 | 24.38 | ↑ higher | Source ↗official | |
| SALAD-Benchmcq_socioeconomic_harms | #20 / 33 | 29.44 | ↑ higher | Source ↗official | |
| SORRY-Benchavg | #2 / 51 | 0.07 | ↓ lower | Source ↗official | |
| SuperCLUE Safetyinstruction_attack | #8 / 31 | 70.69 | ↑ higher | Source ↗official | |
| SuperCLUE Safetyresponsible_ai | #1 / 31 | 78.18 | ↑ higher | Source ↗official | |
| SuperCLUE Safetytraditional_safety | #14 / 31 | 77.66 | ↑ higher | Source ↗official |