Model profile
Zephyr 7B Beta
Evidence summary
Safety. Zephyr 7B Beta has an estimated Safety rank of #293; its 90% source-sensitivity interval is #156–#317. Its behavior-only rank is #303; company governance moves the combined estimate to #293. Published Safety evidence spans 5 eval lineages and 4 of 7 components. Its strongest relative result is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #62 of 248); its weakest is HarmBench (dr, #28 of 28).
Freedom. Zephyr 7B Beta has an estimated Freedom rank of #20; its 90% source-sensitivity interval is #11–#196. Published Freedom evidence spans 4 eval lineages and 1 of 1 components. Its strongest relative result is Enkrypt AI Safety Leaderboard (harmful_attack_non_success_rate, #2 of 248); its weakest is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #184 of 248).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Hugging Face ↗HuggingFaceH4/zephyr-7b-beta
- OpenRouter ↗huggingfaceh4/zephyr-7b-beta
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- Release source ↗direct research preview identity date
Finetunes
Direct finetunes linked to this canonical base model. Quantized and repackaged derivatives are excluded.
- richardyoung/zephyr-7b-beta-abliterated ↗Direct non-quantized finetune documented by Hugging Face metadata
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 |
|---|---|---|---|---|
| Contextual MoralChoicehuman_agreement | #12 / 22 | ↑0.4 | Source ↗official | |
| DecodingTrustmachine_ethics | #6 / 8 | ↑41.03 | Source ↗official | |
| DecodingTruststereotype_bias | #5 / 8 | ↑92.6 | Source ↗official | |
| DecodingTrusttoxicity | #7 / 8 | ↑31.97 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #129 / 248 | ↑15.5 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #62 / 248 | ↑91.17 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #247 / 248 | ↑20 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #174 / 246 | ↑93.94 | Source ↗official | |
| HarmBenchdr | #28 / 28 | ↓65.8 | Source ↗official | |
| SORRY-Benchavg | #50 / 51 | ↓0.87 | 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 |
|---|---|---|---|---|
| DecodingTrusttoxicity | #2 / 8 | ↓31.97 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #184 / 248 | ↓91.17 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #2 / 248 | ↓20 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #73 / 246 | ↓93.94 | Source ↗official | |
| HarmBenchdr | #1 / 28 | ↑65.8 | Source ↗official | |
| SORRY-Benchavg | #2 / 51 | ↑0.87 | Source ↗official |