Model profile
Claude Sonnet 5
Evidence summary
Claude Sonnet 5 has an estimated overall rank of #71; its 90% source-sensitivity interval is #3–#185. Its behavior-only rank is #84; company governance moves the combined estimate to #71. Published evidence spans 10 evals and 7 of 7 behavior components. Its strongest relative result is Enkrypt AI Safety Leaderboard (harmful_attack_non_success_rate, #1 of 260); its weakest is TAC (base_welfare_rate, #67 of 68).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗claude-sonnet-5
- OpenRouter ↗anthropic/claude-sonnet-5
- System card ↗Exact model document · Anthropic · first party
- Release source ↗curated outward evidence
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 |
|---|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #28 / 311 | 0.3725 | ↓ lower | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #5 / 105 | 0.79 | ↑ higher | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #2 / 105 | 2.243 | ↓ lower | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #6 / 260 | 70.54 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #50 / 260 | 92.5 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #1 / 260 | 100 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #14 / 258 | 99.82 | ↑ higher | Source ↗official | |
| Gray Swan indirect prompt injection (15 attempts)attack_success_probability_k15_pct | #5 / 13 | 5.9 | ↓ lower | Source ↗official | |
| Manager Coercion Benchcoercion_ladder_depth | #6 / 31 | 5.5 | ↓ lower | Source ↗official | |
| Manager Coercion Benchfabrication_rate | #1 / 13 | 0 | ↓ lower | Source ↗official | |
| PHAREbias_resistance_diagnostic | #43 / 66 | 0.3988 | ↑ higher | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #7 / 70 | 0.861 | ↑ higher | Source ↗official | |
| PHAREharm_resistance_diagnostic | #4 / 70 | 0.9959 | ↑ higher | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #1 / 67 | 0.8684 | ↑ higher | Source ↗official | |
| SM-Benchadversarial | #39 / 73 | 80.98 | ↑ higher | Source ↗official | |
| SM-Benchambiguous_interpretation | #6 / 73 | 91.96 | ↑ higher | Source ↗official | |
| SM-Benchanti_hallucination | #1 / 73 | 100 | ↑ higher | Source ↗official | |
| SM-Bencheq_boundaries | #58 / 73 | 55.06 | ↑ higher | Source ↗official | |
| SM-Benchoverfit | #18 / 73 | 83.06 | ↑ higher | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #72 / 102 | 6.57 | ↑ higher | Source ↗official | |
| SpeciEvalland_animal_4ns | #78 / 102 | 4.8 | ↓ lower | Source ↗official | |
| SpeciEvalsea_animal_4ns | #84 / 102 | 5.05 | ↓ lower | Source ↗official | |
| SpeciEvalspeciesism | #54 / 102 | 2.07 | ↓ lower | Source ↗official | |
| TACbase_welfare_rate | #67 / 68 | 15.38 | ↑ higher | Source ↗official |
Values evaluations
Descriptive values and political-framing results are separate from safety/ethics ranks. Each strip shows the evaluation’s observed model range; its endpoint labels state what lower and higher values mean.