GLM-4 ranks first on Needle In A Haystack.
Closed weights Flagship
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 | |
|---|---|---|---|---|
| Needle In A Haystack | 100% | #1 of 9 | Lab model card | |
| BIG-Bench Hard | 82.3% | #32 of 73 | Lab model card | |
| LongBench | 81.1% | #1 of 2 | Lab model card | |
| HellaSwag | 85.4% | #27 of 53 | Lab model card | |
| MMLU | 81.5% | #67 of 131 | Lab model card | |
| AlignBench | 7.88 | #12 of 15 | Lab model card | |
| IFEval | 67.7% | #126 of 135 | Lab model card | |
| GLM-4 All Tools information seeking (Z.ai) | 78.1% | #1 of 1 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Math23K | 88.5% | #1 of 3 | Lab model card | |
| MathBench | 51.3% | #10 of 21 | Official leaderboard | |
| GSM8K | 87.6% | #52 of 92 | Lab model card | |
| MATH | 47.9% | #74 of 99 | Lab model card | |
| GPQA Diamond | 35.7% | #283 of 302 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CodeEditorBench | 75% | #3 of 7 | Official leaderboard | |
| HumanEval | 72% | #71 of 101 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AgentBench | 2.89 | #3 of 5 | Official leaderboard |
The Self-Reported Index fits one score to the 9 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 113.5 to 119.9. How we compute it.
The benchmarks the lab chose to headline, with the scores it gave.
Every result in the longer documents the lab published with it. 3 of them aren't in the launch post.