- Terminal Bench 2.1 (Claude Code 2.1.207) 84.3
- DeepSWE (v1.1) 63.4
- NL2Repo 56.3
- Toolathlon Verified 78.4
- AutomationBench (v1.0.6) 48.8
- Agents Last Exam 26.3
- HLE w/ Tools (full set) 55.3
- GDPval-AA v2 1773
- OfficeQA Pro (Treasury Bulletin PDFs, no embedded text) 62.4
- CharXiv Reasoning (w/ Tools) 89.4
- Chartography (w/ Tools) 78
- BabyVision 53.4
- MVbench 77.8
- MMVU 80.5
- OSWorld 2.0 59.1
- Vision2Web 77.8
GLM-5.3-Flash
GLM-5.3-Flash ranks first on 2 of the 100 benchmarks it has scores on.
Z.ai Released 20 Aug 2026
100
Benchmarks with a score
2
Ranked first, out of five or more models
5
In the top three
Z.ai Aug 2026
GLM-5.3-Flash
Open weights
153.2
#48 of 271
Epoch Capabilities Index
- DeepSWE 63.4% #13 /28
- NL2Repo-Bench 56.3% #8 /28
- Terminal-Bench 4.0 35.8% #12 /23
Where it ranks
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.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LMArena (Elo) | 1,477 | #29 of 333 | Official leaderboard | |
| Humanity's Last Exam | 55.3% | #23 of 147 | Model card | |
| AgentIF (2026-10 setting) | 64.4% | #8 of 24 | Official leaderboard | |
| FACTS Grounding v2 | 56.1% | #14 of 15 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GPQA Diamond | 90.2% | #43 of 219 | Epoch AI | |
| OTIS Mock AIME 2024-2025 | 93.9% | #48 of 202 | Epoch AI | |
| CritPt | 15.4% | #44 of 160 | Epoch AI | |
| ARC-AGI-2 | 65.8% | #28 of 95 | Official leaderboard | |
| Verifying the Verifiers (Project APE CRED benchmark) | 90% | #7 of 23 | Official leaderboard | |
| ARC-AGI | 91% | #29 of 92 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 55.8% | #49 of 89 | Epoch AI | |
| NatureBench | 11.1% | #10 of 18 | Official leaderboard | |
| Surface Evolver Bench | 52.5% | #15 of 26 | Epoch AI | |
| ProofBench | 21% | #51 of 83 | Epoch AI | |
| FrontierMath Tier 4 v2 | 17.1% | #54 of 71 | Epoch AI | |
| ProofBench v1.1 | 21% | #40 of 51 | Official leaderboard | |
| BTF-3 | 0.15 | #10 of 10 | Epoch AI |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| WebDev Arena | 1,607 | #19 of 105 | Epoch AI | |
| Code Arena | 1,607 | #27 of 124 | Official leaderboard | |
| NL2Repo-Bench | 56.3% | #8 of 28 | Model card | |
| StructEval | 76.2% | #10 of 28 | Official leaderboard | |
| Image-to-WebDev Arena | 1,578 | #21 of 55 | Official leaderboard | |
| Vision2Web | 77.8% | #4 of 10 | Official leaderboard | |
| PostTrainBench 1.2 | 36.8% | #7 of 17 | Official leaderboard | |
| DeepSWE | 63.4% | #13 of 28 | Official leaderboard | |
| Code Migration v1.1 | 51.6% | #15 of 29 | Official leaderboard | |
| IOI v2 (Vals) | 52.5% | #24 of 45 | Official leaderboard | |
| Vibe Code Bench | 30.8% | #74 of 115 | Official leaderboard | |
| Code Migration | 20.5% | #52 of 78 | Official leaderboard | |
| FrontierCode | 31.8% | #31 of 46 | Epoch AI | |
| Vibe Code Bench 1-100 | 16% | #18 of 26 | Official leaderboard | |
| FrontierSWE | 18.1% | #17 of 21 | Epoch AI | |
| SWE-Marathon v1.1 | 38.8% | #14 of 17 | Official leaderboard | |
| CursorBench | 36.8% | #16 of 19 | Epoch AI | |
| FrontierCS 2 | 12.5% | #14 of 16 | Official leaderboard | |
| FrontierCode 2.0 | 14.2% | #23 of 26 | Official leaderboard | |
| ALE-Bench | 304 | #112 of 121 | Epoch AI | |
| Brokk Power Ranking | 1.60 | #18 of 19 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Toolathlon | 78.4% | #1 of 37 | Model card | |
| Toolathlon-Verified | 78.4% | #1 of 25 | Official leaderboard | |
| AutomationBench | 48.8% | #11 of 41 | Model card | |
| LLM Chess | 572 | #38 of 124 | Official leaderboard | |
| Terminal-Bench 2.1 | 84.3% | #17 of 53 | Model card | |
| Terminal-Bench 4.0 | 35.8% | #12 of 23 | Official leaderboard | |
| OSWorld 2.0 | 59.1% | #10 of 19 | Model card | |
| GAIA | 43.9% | #33 of 57 | Official leaderboard | |
| Agent Arena | -0.02 | #34 of 53 | Official leaderboard | |
| Chess Puzzles | 21% | #62 of 83 | Epoch AI | |
| CivBench | 1,468 | #14 of 18 | Official leaderboard | |
| Hermes Bench | 68.9% | #14 of 17 | Official leaderboard | |
| Mystery Game Puzzles | 8.00% | #71 of 80 | Epoch AI |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MedScribe | 88.9% | #9 of 98 | Official leaderboard | |
| GDPval-AA v2 | 1,773 | #3 of 26 | Model card | |
| TaxEval v2 | 75.6% | #16 of 129 | Official leaderboard | |
| Vals Finance Agent v2 | 57.9% | #14 of 80 | Official leaderboard | |
| Legal Research Bench | 45.2% | #19 of 88 | Official leaderboard | |
| Tax Agent Bench | 31.6% | #24 of 79 | Official leaderboard | |
| LegalBench | 83.9% | #45 of 138 | Official leaderboard | |
| Vals Legal Index | 71% | #13 of 33 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 6.67% | #32 of 80 | Official leaderboard | |
| BlenderWorld | 59.7 | #6 of 13 | Official leaderboard | |
| APEX-Agents | 52.8% | #26 of 56 | Epoch AI | |
| Vals Finance Agent v2.1 | 60.6% | #20 of 35 | Official leaderboard | |
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 10.4% | #19 of 33 | Official leaderboard | |
| E-Commerce Bench | 2.58 | #11 of 19 | Official leaderboard | |
| Vals Finance Index | 55.8% | #22 of 35 | Official leaderboard | |
| Excel Modeling Benchmark | 55.9% | #48 of 76 | Official leaderboard | |
| Clipping Bench | $399 | #13 of 19 | Official leaderboard | |
| OfficeQA Pro | 62.4% | #10 of 14 | Model card | |
| Excel Modeling Benchmark v1.1 | 66.3% | #27 of 37 | Official leaderboard | |
| Thrive Holdings helpdesk ticket resolution benchmark | 42.3% | #11 of 15 | Official leaderboard | |
| Agents' Last Exam | 26.3% | #19 of 25 | Model card | |
| Holdings-IndividualTaxBench | 71.3% | #10 of 13 | Official leaderboard | |
| Vals Index v2.2 | 50.9% | #26 of 32 | Official leaderboard | |
| Legora BAR | 94.0 | #15 of 17 | Official leaderboard | |
| OR-License | 62.6% | #9 of 10 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MVBench | 77.8% | #2 of 15 | Model card | |
| CharXiv | 89.4% | #9 of 67 | Model card | |
| MMVU | 80.5% | #3 of 20 | Model card | |
| ParseBench | 70.8% | #8 of 42 | Official leaderboard | |
| Vision Arena | 1,271 | #29 of 136 | Official leaderboard | |
| Chartography | 78% | #5 of 12 | Model card | |
| VoxelBench | 1,726 | #29 of 56 | Official leaderboard | |
| Humanity's Sixth Sense | 26.8% | #20 of 29 | Official leaderboard | |
| BabyVision | 53.4% | #14 of 20 | Model card | |
| DataCenterBench Tier 1 | 15% | #7 of 9 | Official leaderboard | |
| Code2Video Benchmark | 1,467 | #15 of 19 | Official leaderboard | |
| ScienceVision | 0.28 | #9 of 11 | Official leaderboard | |
| Embodied First Exam | 2.63% | #6 of 7 | Official leaderboard | |
| GDP.pdf | 14% | #34 of 39 | Epoch AI | |
| VibeVideoBench | 2.79 | #9 of 9 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| DistressBench | 72.7% | #18 of 42 | Official leaderboard | |
| CyberBench (Vals) | 71.5% | #12 of 26 | Official leaderboard | |
| CyberBench v1.1 (Vals) | 67.9% | #23 of 43 | Official leaderboard | |
| DeepsecBench | 12.9 | #21 of 28 | Official leaderboard | |
| SRE-Bench | 0.76% | #26 of 33 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SAGE v1.1 (Vals) | 89.3% | #5 of 28 | Official leaderboard | |
| SuperCLUE | 68.1 | #7 of 18 | Official leaderboard | |
| LiveBench | 62.1% | #20 of 28 | Official leaderboard | |
| Hermes Index | 35.0 | #13 of 17 | Official leaderboard |
Z.ai's models over time
Z.ai's models in color, against every other model on the Epoch Capabilities Index, by release date.
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.