← Models

Model profile

Sarvam 105B

Sarvam AIdeveloper
2026-03-03release date
Not rankedSafety rank
#658 / 662Freedom rank

Evidence summary

Safety. Sarvam 105B does not meet the evidence gate for a Safety rank. Published Safety evidence spans 2 eval lineages and 4 of 7 components. Its strongest relative result is Enkrypt AI Safety Leaderboard (harmful_attack_non_success_rate, #44 of 270); its weakest is AA-Omniscience (hallucination_rate, #330 of 358).

Freedom. Sarvam 105B has an estimated Freedom rank of #658; its 90% source-sensitivity interval is #537–#655. Published Freedom evidence spans 1 eval lineages and 1 of 1 components. Its strongest relative result is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #195 of 270); its weakest is Enkrypt AI Safety Leaderboard (harmful_attack_non_success_rate, #225 of 270).

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#330 / 358↓0.9343Source ↗official
Enkrypt AI Safety Leaderboardbias_attack_non_success_rate#247 / 270↑9.04Source ↗official
Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate#75 / 270↑90.5Source ↗official
Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate#44 / 270↑96.11Source ↗official
Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate#56 / 268↑98.36Source ↗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-evalRankValueDistributionSource
Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate#195 / 270↓90.5Source ↗official
Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate#225 / 270↓96.11Source ↗official
Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate#211 / 268↓98.36Source ↗official