GLM-4.6 places in the top three on 2 benchmarks.
Open 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 | |
|---|---|---|---|---|
| ConStory-Bench | 0.53 | #3 of 14 | Official leaderboard | |
| LMArena (Elo) | 1,424 | #100 of 335 | Official leaderboard | |
| Encyclo-K | 36.5% | #9 of 23 | Official leaderboard | |
| Creative Writing v3 | 69.3 | #58 of 120 | Official leaderboard | |
| Vectara Hallucination Leaderboard (HHEM) | 4.30% | #75 of 148 | Official leaderboard | |
| Vectara Hallucination Leaderboard (2025 dataset) | 9.50% | #54 of 104 | Official leaderboard | |
| Humanity's Last Exam | 30.4% | #80 of 154 | Lab model card | |
| Longform Creative Writing (EQ-Bench) | 57.3 | #73 of 119 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| HMMT 2025 | 93.3% | #8 of 53 | Official leaderboard | |
| AIME 2025 | 93.9% | #29 of 143 | Lab model card | |
| SMT 2025 | 90.6% | #9 of 40 | Official leaderboard | |
| BRUMO 2025 | 94.2% | #14 of 42 | Official leaderboard | |
| CMIMC 2025 | 87.5% | #11 of 33 | Official leaderboard | |
| HMMT Nov 2025 | 91.7% | #8 of 22 | Official leaderboard | |
| GPQA Diamond | 81% | #112 of 302 | Lab model card | |
| General365 | 52.9% | #14 of 23 | Official leaderboard | |
| FrontierMath Tier 4 | 2.13% | #36 of 55 | Epoch AI run | |
| FrontierMath | 3.82% | #45 of 55 | Official leaderboard | |
| MathArena (Apex, IMO, USAMO) | 0.52% | #45 of 47 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BigCodeArena | 1,106 | #2 of 13 | Official leaderboard | |
| SWE-bench Verified | 68.2% | #17 of 60 | Official leaderboard | |
| LiveCodeBench v6 | 82.8% | #26 of 90 | Lab model card | |
| DesignArena Website | 1,167 | #92 of 151 | Official leaderboard | |
| DesignArena Game Dev | 1,141 | #88 of 143 | Official leaderboard | |
| DesignArena Code | 1,156 | #92 of 143 | Official leaderboard | |
| DesignArena 3D Design | 1,111 | #88 of 136 | Official leaderboard | |
| ts-bench v1 | 92% | #6 of 9 | Official leaderboard | |
| DesignArena UI Component | 1,139 | #94 of 139 | Official leaderboard | |
| DesignArena SVG | 1,099 | #74 of 103 | Official leaderboard | |
| DesignArena Data Visualization | 1,159 | #86 of 118 | Official leaderboard | |
| SWE-fficiency | 4.20% | #13 of 17 | Official leaderboard | |
| WebDev Arena | 1,341 | #83 of 105 | arena.ai, via Epoch AI | |
| IOI (Vals) | 4.33% | #44 of 55 | Official leaderboard | |
| Code Arena | 1,342 | #104 of 128 | Official leaderboard | |
| DesignArena React Native | 1,090 | #46 of 56 | Official leaderboard | |
| Vibe Code Bench v1 | 3.09% | #14 of 17 | Official leaderboard | |
| DesignArena Agentic Overall | 1,052 | #61 of 74 | Official leaderboard | |
| DesignArena Android (Kotlin) | 1,072 | #38 of 46 | Official leaderboard | |
| DesignArena Full Stack | 1,008 | #44 of 49 | Official leaderboard | |
| Vibe Code Bench | 3.09% | #112 of 120 | Official leaderboard | |
| ALE-Bench | 838 | #119 of 126 | Official leaderboard | |
| SWE-Bench Pro | 9.67% | #23 of 24 | Official leaderboard | |
| AInsteinBench | 25.9% | #6 of 6 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BFCL V4 | 72.4% | #4 of 65 | Official leaderboard | |
| SnakeBench | 25 | #77 of 188 | Official leaderboard | |
| MCP-Universe | 26% | #15 of 30 | Official leaderboard | |
| Step Race | 2 | #31 of 59 | Official leaderboard | |
| tau2-bench | 75.9% | #49 of 79 | Lab model card | |
| AgencyBench | 38.6% | #6 of 9 | Official leaderboard | |
| VitaBench 2.0 | 35.9% | #12 of 17 | Official leaderboard | |
| Terminal-Bench | 24.5% | #13 of 18 | tbench.ai, via Epoch AI | |
| ClawProBench | 56.3% | #36 of 49 | Official leaderboard | |
| LisanBench | 157.0 | #98 of 133 | Official leaderboard | |
| Toolathlon | 18.8% | #39 of 50 | Official leaderboard | |
| BrowseComp | 45.1% | #66 of 81 | Lab model card | |
| Terminal-Bench 2.0 | 24.5% | #38 of 43 | Official leaderboard | |
| Elimination Game | 3 | #50 of 56 | Official leaderboard | |
| Agent Poker Bench | 946 | #16 of 17 | Official leaderboard | |
| WorldTravel | 8.00% | #7 of 7 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| PLawBench | 60.5% | #6 of 11 | Official leaderboard | |
| Vals Finance Agent v1 | 27.1% | #27 of 47 | Official leaderboard | |
| LegalBench | 79.6% | #90 of 138 | Official leaderboard | |
| CorpFin v2 | 56.8% | #83 of 119 | Official leaderboard | |
| TaxEval v2 | 66.2% | #105 of 129 | Official leaderboard | |
| Vals Finance Agent | 36.5% | #42 of 49 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MolecularIQ | 16.2% | #4 of 17 | Official leaderboard | |
| nocot-bench (NCRI) | 70.8 | #94 of 185 | Official leaderboard | |
| LiveSecBench | 44.9 | #20 of 34 | Official leaderboard | |
| Spiral-Bench v1.2 | 30.8 | #15 of 21 | Official leaderboard | |
| HelpBench | 80% | #7 of 9 | Official leaderboard |
The Self-Reported Index fits one score to the 8 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 140.0 to 144.4. 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.
The benchmarks the lab chose to headline, with the scores it gave.