GLM-4.5-Air places in the top three on SysBench.
Open weights
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Vectara Hallucination Leaderboard (HHEM) | 1.90% | #23 of 148 | Official leaderboard | |
| MMLU | 87.4% | #36 of 131 | Lab model card | |
| MMLU-Pro | 81.4% | #66 of 167 | Lab model card | |
| Vectara Hallucination Leaderboard (2025 dataset) | 9.30% | #49 of 104 | Official leaderboard | |
| LMArena (Elo) | 1,373 | #160 of 335 | Official leaderboard | |
| SysBench | 77.4% | #3 of 6 | Lab model card | |
| IFEval | 86.3% | #69 of 135 | Lab model card | |
| LongBench v2 | 48.6% | #18 of 33 | Official leaderboard | |
| Wer wird Millionär? LLM Benchmark | 1867.0 | #45 of 77 | Official leaderboard | |
| Humanity's Last Exam | 8.12% | #33 of 47 | Official leaderboard | |
| SimpleQA | 14.5% | #43 of 60 | Lab model card | |
| MultiChallenge | 42.5% | #36 of 42 | Lab model card | |
| MultiNRC | 10.4% | #36 of 38 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MATH-500 | 98.1% | #5 of 70 | Lab model card | |
| AIME 2024 | 89.4% | #21 of 88 | Lab model card | |
| GPQA Main | 59.4% | #26 of 60 | Official leaderboard | |
| GPQA Diamond | 75% | #150 of 302 | Lab model card | |
| BRUMO 2025 | 90% | #23 of 42 | Official leaderboard | |
| AIME 2025 | 83.3% | #31 of 53 | Official leaderboard | |
| HMMT 2025 | 69.2% | #31 of 53 | Official leaderboard | |
| ForecastBench | 0.15 | #59 of 82 | Official leaderboard | |
| CMIMC 2025 | 71.9% | #25 of 33 | Official leaderboard | |
| SMT 2025 | 77.4% | #32 of 40 | Official leaderboard | |
| Novel logical reasoning problems expert evaluation (Z.ai) | 53.4 | #2 of 2 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveCodeBench v5 | 70.7% | #11 of 47 | Lab model card | |
| ArtifactsBench | 48.9% | #11 of 20 | Official leaderboard | |
| DesignArena Data Visualization | 1,184 | #71 of 118 | Official leaderboard | |
| AutoCodeBench | 40.8% | #22 of 34 | Official leaderboard | |
| DesignArena 3D Design | 1,105 | #92 of 136 | Official leaderboard | |
| DesignArena Website | 1,139 | #105 of 151 | Official leaderboard | |
| DesignArena UI Component | 1,114 | #101 of 139 | Official leaderboard | |
| DesignArena Code | 1,128 | #105 of 143 | Official leaderboard | |
| DesignArena Game Dev | 1,084 | #105 of 143 | Official leaderboard | |
| SciCode | 37.3% | #36 of 49 | Lab model card | |
| SWE-bench Verified | 57.6% | #116 of 155 | Lab model card | |
| DesignArena SVG | 1,062 | #80 of 103 | Official leaderboard | |
| GSO-Bench | 2.94% | #28 of 30 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Berkeley Function Calling Leaderboard (BFCL) | 76.4% | #5 of 41 | Lab model card | |
| Galileo Agent Leaderboard v2 | 44% | #9 of 21 | Official leaderboard | |
| PinchBench | 69.4% | #24 of 48 | Official leaderboard | |
| SnakeBench | 24 | #97 of 188 | Official leaderboard | |
| LisanBench | 238.7 | #85 of 133 | Official leaderboard | |
| ClawBench V2 | 2.30% | #5 of 7 | Official leaderboard | |
| MCP-Universe | 19.5% | #26 of 30 | Official leaderboard | |
| Terminal-Bench | 30% | #21 of 22 | Lab model card | |
| BrowseComp | 21.3% | #81 of 81 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SimpleSafetyTests | 99% | #38 of 75 | Official leaderboard | |
| Anthropic Red Team | 99% | #39 of 75 | Official leaderboard | |
| HELM Safety | 90.1 | #47 of 75 | Official leaderboard | |
| AIR-Bench 2024 | 57.1% | #56 of 82 | Official leaderboard | |
| nocot-bench (NCRI) | 62.4 | #133 of 185 | Official leaderboard | |
| FORTRESS | 63.2% | #72 of 73 | Official leaderboard | |
| SafetyBench | 87.8% | #2 of 2 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| HELM Capabilities | 67.0 | #28 of 60 | Official leaderboard | |
| Artificial Analysis Intelligence Index v2 | 64.8 | #2 of 2 | Lab model card |
The Self-Reported Index fits one score to the 15 capability results Z.ai reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 132.1 to 136.7. How we compute it.
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
Every result in the longer documents the lab published with it. 20 of them aren't in the launch post.