Model profile
Kimi K2.5
Moonshot AIdeveloper
2026-01-01release date
#85 / 267overall rank
22eval lineages
Evidence summary
Kimi K2.5 has an estimated overall rank of #85; its 90% source-sensitivity interval is #52–#167. Its behavior-only rank is #83; company governance moves the combined estimate to #85. Published evidence spans 22 evals and 7 of 7 behavior components. Its strongest relative result is LiveSecBench (factuality, #4 of 43); its weakest is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #251 of 260).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗kimi-k2-5
- Hugging Face ↗moonshotai/Kimi-K2.5
- OpenRouter ↗moonshotai/kimi-k2.5
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- Release source ↗direct research preview identity date
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 | #44 / 311 | 0.4786 | ↓ lower | Source ↗official | |
| AgentAbstainabstain | #13 / 17 | 52 | ↑ higher | Source ↗official | |
| AgentAbstaincar | #11 / 17 | 53.2 | ↑ higher | Source ↗official | |
| AgentAbstainpaired | #16 / 17 | 33.4 | ↑ higher | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #35 / 105 | 0.415 | ↑ higher | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #36 / 43 | 90.9 | ↓ lower | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #42 / 48 | 93.2 | ↓ lower | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #43 / 49 | 75.9 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmachiavelli | #36 / 45 | 93.5 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmask | #24 / 51 | 18.5 | ↓ lower | Source ↗official | |
| CAIS Risk Indextextquests_harm | #34 / 48 | 20.8 | ↓ lower | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #42 / 105 | 24.13 | ↓ lower | Source ↗official | |
| DystopiaBenchbasaglia_score | #20 / 50 | 59.1 | ↓ lower | Source ↗official | |
| DystopiaBenchbaudrillard_score | #17 / 50 | 44.77 | ↓ lower | Source ↗official | |
| DystopiaBenchhuxley_score | #20 / 50 | 62.6 | ↓ lower | Source ↗official | |
| DystopiaBenchlaguardia_score | #20 / 50 | 62.37 | ↓ lower | Source ↗official | |
| DystopiaBenchorwell_score | #19 / 50 | 56.97 | ↓ lower | Source ↗official | |
| DystopiaBenchpetrov_score | #24 / 50 | 72.53 | ↓ lower | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #123 / 260 | 16.28 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #251 / 260 | 66.83 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #37 / 260 | 94.44 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #128 / 258 | 96.36 | ↑ higher | Source ↗official | |
| FORTRESSaverage_risk_score | #29 / 49 | 41.09 | ↓ lower | Source ↗official | |
| FORTRESSover_refusal_score | #20 / 46 | 3.72 | ↓ lower | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #12 / 54 | 28.21 | ↑ higher | Source ↗official | |
| LiveSecBenchethics | #7 / 43 | 79.3 | ↑ higher | Source ↗official | |
| LiveSecBenchfactuality | #4 / 43 | 81.18 | ↑ higher | Source ↗official | |
| LiveSecBenchlegality | #13 / 43 | 67.47 | ↑ higher | Source ↗official | |
| LiveSecBenchprivacy | #17 / 43 | 60.89 | ↑ higher | Source ↗official | |
| LiveSecBenchpsychological_health | #4 / 43 | 85.12 | ↑ higher | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #37 / 50 | 93.5 | ↓ lower | Source ↗official | |
| Manager Coercion Benchcoercion_ladder_depth | #26 / 31 | 8.967 | ↓ lower | Source ↗self run | |
| MASKlying_probability_pct | #21 / 53 | 29.53 | ↓ lower | Source ↗official | |
| PHAREbias_resistance_diagnostic | #62 / 66 | 0.29 | ↑ higher | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #18 / 70 | 0.8069 | ↑ higher | Source ↗official | |
| PHAREharm_resistance_diagnostic | #10 / 70 | 0.972 | ↑ higher | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #23 / 67 | 0.613 | ↑ higher | Source ↗official | |
| SABERoverall_safety_rate | #8 / 13 | 23.9 | ↑ higher | Source ↗official | |
| SABERscenario_a_safety_rate | #6 / 13 | 28.95 | ↑ higher | Source ↗official | |
| SABERscenario_b_safety_rate | #9 / 13 | 28.24 | ↑ higher | Source ↗official | |
| SABERscenario_c_safety_rate | #9 / 13 | 14.48 | ↑ higher | Source ↗official | |
| SM-Benchadversarial | #22 / 73 | 84.39 | ↑ higher | Source ↗official | |
| SM-Benchambiguous_interpretation | #15 / 73 | 89.29 | ↑ higher | Source ↗official | |
| SM-Benchanti_hallucination | #34 / 73 | 93.19 | ↑ higher | Source ↗official | |
| SM-Bencheq_boundaries | #51 / 73 | 58.15 | ↑ higher | Source ↗official | |
| SM-Benchoverfit | #40 / 73 | 68.85 | ↑ higher | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #50 / 102 | 6.8 | ↑ higher | Source ↗official | |
| SpeciEvalland_animal_4ns | #20 / 102 | 4.28 | ↓ lower | Source ↗official | |
| SpeciEvalsea_animal_4ns | #12 / 102 | 4.33 | ↓ lower | Source ↗official | |
| SpeciEvalspeciesism | #28 / 102 | 1.73 | ↓ lower | Source ↗official | |
| TACbase_welfare_rate | #42 / 68 | 24.36 | ↑ higher | Source ↗self run | |
| ToolPrivacyBenchprivate_mt_poi | #8 / 9 | 27.74 | ↓ lower | Source ↗official | |
| ToolPrivacyBenchpublic_mt_poi | #1 / 9 | 15.81 | ↓ lower | 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.
UGI Political Values
Taiwan Sovereignty Benchmark Pro
| Dimension | Value | Distribution |
|---|---|---|
| Pro-Taiwan rubric compatibility | 15 | |
| Warning-phrase rate | 40 | |
| Soft-censorship rate | 0 | |
| API-error rate | 0 |
Agent-ValueBench Moral Foundations (MFT08)
Agent-ValueBench HEXACO
| Dimension | Value | Distribution |
|---|---|---|
| Openness to experience | 5.8 | |
| Honesty-humility | 6 | |
| Extraversion | 6.7 | |
| Agreeableness | 6.8 | |
| Conscientiousness | 7.3 |