Model profile
Kimi K2.6
Evidence summary
Kimi K2.6 has an estimated overall rank of #63; its 90% source-sensitivity interval is #37–#151. Its behavior-only rank is #61; company governance moves the combined estimate to #63. Published evidence spans 18 evals and 7 of 7 behavior components. Its strongest relative result is AA-Omniscience (hallucination_rate, #32 of 311); its weakest is DystopiaBench (petrov_score, #46 of 50).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗kimi-k2-6
- Hugging Face ↗moonshotai/Kimi-K2.6
- NVIDIA ↗qc69jvmznzxy/kimi-k2.6
- OpenRouter ↗moonshotai/kimi-k2.6
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- Release source ↗direct research preview identity date
- www.kimi.com ↗
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 | #32 / 311 | 0.3926 | ↓ lower | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #17 / 105 | 0.575 | ↑ higher | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #34 / 43 | 90.7 | ↓ lower | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #37 / 48 | 88.4 | ↓ lower | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #28 / 49 | 58.2 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmachiavelli | #30 / 45 | 92.1 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmask | #28 / 51 | 23.2 | ↓ lower | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #12 / 32 | 43.6 | ↓ lower | Source ↗official | |
| CAIS Risk Indextextquests_harm | #31 / 48 | 20.4 | ↓ lower | Source ↗official | |
| DystopiaBenchbasaglia_score | #24 / 50 | 62.57 | ↓ lower | Source ↗official | |
| DystopiaBenchbaudrillard_score | #19 / 50 | 46.73 | ↓ lower | Source ↗official | |
| DystopiaBenchhuxley_score | #21 / 50 | 63.03 | ↓ lower | Source ↗official | |
| DystopiaBenchlaguardia_score | #12 / 50 | 53.97 | ↓ lower | Source ↗official | |
| DystopiaBenchorwell_score | #22 / 50 | 62.3 | ↓ lower | Source ↗official | |
| DystopiaBenchpetrov_score | #46 / 50 | 82.73 | ↓ lower | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #18 / 54 | 27.72 | ↑ higher | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #31 / 50 | 92.1 | ↓ lower | Source ↗official | |
| Manager Coercion Benchcoercion_ladder_depth | #22 / 31 | 8.9 | ↓ lower | Source ↗self run | |
| ODCV-Benchaverage_severity | #5 / 12 | 1.337 | ↓ lower | Source ↗official | |
| ODCV-Benchmisalignment_rate | #7 / 12 | 33.75 | ↓ lower | Source ↗official | |
| PHAREbias_resistance_diagnostic | #19 / 66 | 0.5307 | ↑ higher | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #11 / 70 | 0.8278 | ↑ higher | Source ↗official | |
| PHAREharm_resistance_diagnostic | #11 / 70 | 0.97 | ↑ higher | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #10 / 67 | 0.714 | ↑ higher | Source ↗official | |
| RefusalBenchyouden_j | #15 / 19 | 0.08511 | ↑ higher | Source ↗official | |
| SM-Benchadversarial | #24 / 73 | 83.9 | ↑ higher | Source ↗official | |
| SM-Benchambiguous_interpretation | #13 / 73 | 89.88 | ↑ higher | Source ↗official | |
| SM-Benchanti_hallucination | #22 / 73 | 96.34 | ↑ higher | Source ↗official | |
| SM-Bencheq_boundaries | #45 / 73 | 60.67 | ↑ higher | Source ↗official | |
| SM-Benchoverfit | #41 / 73 | 67.21 | ↑ higher | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #36 / 102 | 6.87 | ↑ higher | Source ↗official | |
| SpeciEvalland_animal_4ns | #33 / 102 | 4.38 | ↓ lower | Source ↗official | |
| SpeciEvalsea_animal_4ns | #16 / 102 | 4.47 | ↓ lower | Source ↗official | |
| SpeciEvalspeciesism | #34 / 102 | 1.8 | ↓ lower | Source ↗official | |
| TACbase_welfare_rate | #39 / 68 | 25.64 | ↑ higher | Source ↗self run |
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.