Model profile
Magistral Small
Evidence summary
Safety. Magistral Small has an estimated Safety rank of #342; its 90% source-sensitivity interval is #295–#345. Its behavior-only rank is #335; company governance moves the combined estimate to #342. Published Safety evidence spans 5 eval lineages and 5 of 7 components. Its strongest relative result is Human Pathogen Capabilities Test (HPCT) — overall refusal (refusal_rate_pct, #27 of 69); its weakest is PHARE (harm_resistance_diagnostic, #68 of 70).
Freedom. Magistral Small has an estimated Freedom rank of #1; its 90% source-sensitivity interval is #4–#27. Published Freedom evidence spans 5 eval lineages and 1 of 1 components. Its strongest relative result is Human Pathogen Capabilities Test (HPCT) — overall refusal (refusal_rate_pct, #1 of 69); its weakest is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #72 of 270).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗magistral-small
- Hugging Face ↗mistralai/Magistral-Small-2506
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- Release source ↗repository created at proxy
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-eval | Rank | Value | Distribution | Source |
|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #216 / 358 | ↓0.8345 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #94 / 104 | ↓74.28 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #129 / 270 | ↑16.54 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #199 / 270 | ↑80.34 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #247 / 270 | ↑34.72 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #231 / 268 | ↑89.59 | Source ↗official | |
| Human Pathogen Capabilities Test (HPCT) — overall refusalrefusal_rate_pct | #27 / 69 | ↑0 | Source ↗official | |
| PHAREbias_resistance_diagnostic | #28 / 66 | ↑0.4802 | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #64 / 70 | ↑0.6244 | Source ↗official | |
| PHAREharm_resistance_diagnostic | #68 / 70 | ↑0.7623 | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #57 / 67 | ↑0.3709 | Source ↗official |
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-eval | Rank | Value | Distribution | Source |
|---|---|---|---|---|
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #11 / 104 | ↑74.28 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #72 / 270 | ↓80.34 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #24 / 270 | ↓34.72 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #38 / 268 | ↓89.59 | Source ↗official | |
| Human Pathogen Capabilities Test (HPCT) — overall refusalrefusal_rate_pct | #1 / 69 | ↓0 | Source ↗official | |
| PHAREharm_resistance_diagnostic | #3 / 70 | ↓0.7623 | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #11 / 67 | ↓0.3709 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_adherence_score | #20 / 156 | ↑6.75 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_direct_score | #3 / 156 | ↑6.5 | 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.