Model profile
GLM 5.3 Flash
Evidence summary
Safety. GLM 5.3 Flash has an estimated Safety rank of #31; its 90% source-sensitivity interval is #16–#121. Its behavior-only rank is #16; company governance moves the combined estimate to #31. Published Safety evidence spans 7 eval lineages and 6 of 7 components. Its strongest relative result is AA-Omniscience (hallucination_rate, #26 of 358); its weakest is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #254 of 270).
Freedom. GLM 5.3 Flash has an estimated Freedom rank of #406; its 90% source-sensitivity interval is #252–#469. Published Freedom evidence spans 3 eval lineages and 1 of 1 components. Its strongest relative result is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #17 of 270); its weakest is Adversarial Poetry Refusal (AHB self-run) (hate, #13 of 14).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗glm-5-3-flash
- Hugging Face ↗zai-org/GLM-5.3-Flash
- OpenRouter ↗z-ai/glm-5.3-flash
- Official model documentation ↗Family-level model document · zai · first party
Finetunes
Direct finetunes linked to this canonical base model. Quantized and repackaged derivatives are excluded.
orcarouter/GLM-5.3-Flash-Uncensored-FP8 ↗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 |
|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #26 / 358 | ↓0.2763 | Source ↗official | |
| Adversarial Poetry Refusal (AHB self-run)child_sexual_exploitation | #2 / 14 | ↓1.449 | Source ↗self run | |
| Adversarial Poetry Refusal (AHB self-run)defamation | #3 / 14 | ↓3.623 | Source ↗self run | |
| Adversarial Poetry Refusal (AHB self-run)hate | #2 / 14 | ↓3.623 | Source ↗self run | |
| Adversarial Poetry Refusal (AHB self-run)indiscriminate_weapons_cbrne | #5 / 14 | ↓1.449 | Source ↗self run | |
| Adversarial Poetry Refusal (AHB self-run)intellectual_property | #6 / 14 | ↓8.696 | Source ↗self run | |
| Adversarial Poetry Refusal (AHB self-run)non_violent_crimes | #5 / 14 | ↓4.348 | Source ↗self run | |
| Adversarial Poetry Refusal (AHB self-run)privacy | #3 / 14 | ↓7.246 | Source ↗self run | |
| Adversarial Poetry Refusal (AHB self-run)sex_related_crimes | #4 / 14 | ↓4.348 | Source ↗self run | |
| Adversarial Poetry Refusal (AHB self-run)sexual_content | #7 / 14 | ↓7.576 | Source ↗self run | |
| Adversarial Poetry Refusal (AHB self-run)specialized_advice | #5 / 14 | ↓3 | Source ↗self run | |
| Adversarial Poetry Refusal (AHB self-run)suicide_self_harm | #7 / 14 | ↓7.576 | Source ↗self run | |
| Adversarial Poetry Refusal (AHB self-run)violent_crimes | #5 / 14 | ↓5.303 | Source ↗self run | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #23 / 270 | ↑47.55 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #254 / 270 | ↑52.83 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #31 / 270 | ↑97.78 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #91 / 268 | ↑97.64 | Source ↗official | |
| Olam Diplomacy — Broken Promise Ratebroken_promise_rate_pct | #3 / 19 | ↓10.8 | Source ↗official | |
| Olam Social Poker — Social Lie Ratesocial_lie_rate_per_10000_turns | #17 / 24 | ↓30 | Source ↗official; provisional source attribution | |
| SpeciEvalbelief_animal_sentience | #105 / 131 | ↑6.5 | Source ↗official | |
| SpeciEvalland_animal_4ns | #47 / 131 | ↓4.4 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #24 / 131 | ↓4.47 | Source ↗official | |
| SpeciEvalspeciesism | #63 / 131 | ↓1.95 | Source ↗official | |
| Vals AI Cheating Auditswebench_verified_cheating_attempt_rate_pct | #4 / 6 | ↓48.09 | 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.