← Models

Model profile

MiMo v2.5 Pro

Xiaomideveloper
2026-04-27release date
#71 / 348Safety rank
#520 / 663Freedom rank
1discovery sources

Evidence summary

Safety. MiMo v2.5 Pro has an estimated Safety rank of #71; its 90% source-sensitivity interval is #20–#219. Its behavior-only rank is #67; company governance moves the combined estimate to #71. Published Safety evidence spans 9 eval lineages and 6 of 7 components. Its strongest relative result is Concordia — SOSBench-Bio (safety_score, #1 of 63); its weakest is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #253 of 272).

Freedom. MiMo v2.5 Pro has an estimated Freedom rank of #520; its 90% source-sensitivity interval is #303–#626. Published Freedom evidence spans 5 eval lineages and 1 of 1 components. Its strongest relative result is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #20 of 272); its weakest is Concordia — SOSBench-Bio (safety_score, #61 of 63).

Compare this model

Only models sharing at least one published sub-eval are listed.

Official and reference links

Safety evals

Rank and direction are specific to the Safety portfolio. Black marks the observed result; the grey dot marks the value implied by the Safety rank. Values stay on each source’s native scale.

Eval / sub-evalRankValueDistributionSource
AA-Omnisciencehallucination_rate#17 / 361↓0.247Source ↗official
Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating#38 / 111↑1445.0Source ↗official
BullshitBench v2clear_pushback_rate#36 / 122↑0.5Source ↗official
Concordia — Agentic-Misalignmentsafety_score#30 / 54↑75.67Source ↗official
Concordia — AirBench-Deceptionsafety_score#10 / 63↑96.3Source ↗official
Concordia — AirBench-Manipulationsafety_score#8 / 56↑98.67Source ↗official
Concordia — AirBench-SecurityRiskssafety_score#23 / 63↑96.4Source ↗official
Concordia — APEsafety_score#22 / 55↑39.3Source ↗official
Concordia — CyberSecEval2-PromptInjectionsafety_score#33 / 63↑83.67Source ↗official
Concordia — DarkBenchsafety_score#4 / 55↑69.09Source ↗official
Concordia — Fortress-Biologicalsafety_score#22 / 54↑72.51Source ↗official
Concordia — Fortress-Chemicalsafety_score#16 / 54↑74.9Source ↗official
Concordia — Fortress-Privacy/Scamssafety_score#11 / 54↑76.57Source ↗official
Concordia — FRT-AirBench-Manipulationsafety_score#15 / 45↑44.67Source ↗official
Concordia — FRT-AirBench-SecurityRiskssafety_score#22 / 45↑29.83Source ↗official
Concordia — FRT-SciKnowEval-BiologicalHarmfulQAsafety_score#9 / 45↑7.667Source ↗official
Concordia — FRT-SOSBench-Chemsafety_score#14 / 45↑63.67Source ↗official
Concordia — MASKsafety_score#24 / 62↑68.86Source ↗official
Concordia — SciKnowEval-BiologicalHarmfulQAsafety_score#27 / 63↑84.85Source ↗official
Concordia — SciKnowEval-ChemicalHarmfulQAsafety_score#36 / 63↑17.4Source ↗official
Concordia — Shutdown-Resistancesafety_score#41 / 53↑88Source ↗official
Concordia — SOSBench-Biosafety_score#1 / 63↑100Source ↗official
Concordia — SOSBench-Chemsafety_score#8 / 63↑95.6Source ↗official
DystopiaBenchbasaglia_score#8 / 50↓33.37Source ↗official
DystopiaBenchbaudrillard_score#20 / 50↓51.07Source ↗official
DystopiaBenchhuxley_score#7 / 50↓26.03Source ↗official
DystopiaBenchlaguardia_score#5 / 50↓37.03Source ↗official
DystopiaBenchorwell_score#6 / 50↓26.03Source ↗official
DystopiaBenchpetrov_score#10 / 50↓42.53Source ↗official
Enkrypt AI Safety Leaderboardbias_attack_non_success_rate#60 / 272↑28.94Source ↗official
Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate#253 / 272↑53Source ↗official
Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate#33 / 272↑97.78Source ↗official
Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate#63 / 270↑98.18Source ↗official
Opposite-Narrator Sycophancysycophancy_rate_pct#11 / 24↓2.6Source ↗official
SM-Benchadversarial#37 / 92↑83.41Source ↗official
SM-Benchambiguous_interpretation#42 / 92↑86.61Source ↗official
SM-Benchanti_hallucination#19 / 92↑98.43Source ↗official
SM-Bencheq_boundaries#33 / 92↑67.98Source ↗official
SM-Benchoverfit#26 / 92↑85.79Source ↗official
TACbase_welfare_rate#68 / 94↑23.7Source ↗self run

Freedom evals

Rank and direction are specific to the Freedom portfolio. Black marks the observed result; the grey dot marks the value implied by the Freedom rank. Values stay on each source’s native scale.

Eval / sub-evalRankValueDistributionSource
Concordia — AirBench-Deceptionsafety_score#53 / 63↓96.3Source ↗official
Concordia — AirBench-Manipulationsafety_score#46 / 56↓98.67Source ↗official
Concordia — AirBench-PoliticalPersuasionsafety_score#37 / 56↓70Source ↗official
Concordia — AirBench-SecurityRiskssafety_score#40 / 63↓96.4Source ↗official
Concordia — Fortress-Biologicalsafety_score#33 / 54↓72.51Source ↗official
Concordia — Fortress-Chemicalsafety_score#39 / 54↓74.9Source ↗official
Concordia — Fortress-Privacy/Scamssafety_score#44 / 54↓76.57Source ↗official
Concordia — FRT-AirBench-Manipulationsafety_score#31 / 45↓44.67Source ↗official
Concordia — FRT-AirBench-SecurityRiskssafety_score#24 / 45↓29.83Source ↗official
Concordia — FRT-SciKnowEval-BiologicalHarmfulQAsafety_score#37 / 45↓7.667Source ↗official
Concordia — FRT-SOSBench-Chemsafety_score#32 / 45↓63.67Source ↗official
Concordia — SciKnowEval-BiologicalHarmfulQAsafety_score#36 / 63↓84.85Source ↗official
Concordia — SciKnowEval-ChemicalHarmfulQAsafety_score#28 / 63↓17.4Source ↗official
Concordia — SOSBench-Biosafety_score#61 / 63↓100Source ↗official
Concordia — SOSBench-Chemsafety_score#56 / 63↓95.6Source ↗official
Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate#20 / 272↓53Source ↗official
Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate#232 / 272↓97.78Source ↗official
Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate#203 / 270↓98.18Source ↗official
SM-Benchadversarial#53 / 92↓83.41Source ↗official
SM-Bencheq_boundaries#33 / 92↑67.98Source ↗official
SM-Benchoverfit#26 / 92↑85.79Source ↗official
SpeechMap model completioncomplete_pct#164 / 181↑32.4Source ↗official
UGI Leaderboard — base-model willingnesswillingness_adherence_score#79 / 156↑1.5Source ↗official
UGI Leaderboard — base-model willingnesswillingness_direct_score#79 / 156↑3Source ↗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

DimensionValueDistribution
Political Lean-15.9
Government47.1
Diplomacy63.3
Economy46.5
Society60.8

CCPBench political narrative alignment

DimensionValueDistribution
CCP-narrative alignment — all questions3.37
CCP-narrative alignment — China topics3.83
CCP-narrative alignment — non-China controls1.98