Up to four models
GLM-5.3 vs GLM-5.2
On the index (GPT-4 = 100), GLM-5.3 scores 172 and GLM-5.2 160. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
GLM-5.3172155–188
WeightsOpen
ReleasedAug 2026
Price in / out$0.077 / $1.4
Context1.05M
Core tests taken5 of 9
GLM-5.2160143–176
WeightsOpen
ReleasedJun 2026
Price in / out$0.034 / $7.0
Context1.05M
Core tests taken6 of 9
GLM-5.3 leads GLM-5.2 by 12 points, ahead on 3 of 4 shared core tests, and costs 4.4× less.
Test by testthird-party results
GLM-5.3GLM-5.2Every result as a table
| Test | GLM-5.3 | GLM-5.2 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 90.9% | 91.9% |
| LiveBench Data Analysis | 70.2% | 73.7% |
| LiveBench IF | 69.3% | 62.3% |
| LiveBench Language | 79.9% | 76.2% |
| LiveBench Reasoning | 85.8% | 78.6% |
| GPQA Diamond (Vals AI) | 88.1% | 85.6% |
| MMLU Pro | 86.8% | 86.7% |
| SimpleQA Verified | 41.0% | 34.2% |
| Maths | ||
| FrontierMath (tiers 1-3) | 68.8% | 59.2% |
| AIME-style maths (OTIS mock) | 91.1% | 86.4% |
| LiveBench Mathematics | 87.9% | 89.8% |
| Chess puzzles | 37.0% | 22.0% |
| FrontierMath (tier 4) | 29.3% | 29.3% |
| Coding | ||
| LiveBench Agentic Coding | 60.9% | 51.8% |
| LiveBench Coding | 79.0% | 79.7% |
| Code Migration | 44.2% | 37.9% |
| LiveCodeBench | 80.5% | 69.5% |
| ProgramBench | 1.5% | 0.5% |
| SWE-bench | 95.4% | 82.8% |
| Vibe Code Bench v1.1 | 78.1% | 64.0% |
| DeepSWE | 69.0% | 43.8% |
| WeirdML | 75.4% | 70.1% |
| Agents | ||
| APEX-Agents | 56.6% | 45.2% |
| SkillsBench | 47.5% | 45.1% |
| SRE Bench | 9.2% | 0.0% |
| DTBench | 87.7% | 93.6% |
| LMCA | 55.5% | 45.8% |
| Mystery game puzzles | 33.0% | 19.0% |
| Professional work | ||
| Finance Agent (v2) | 55.8% | 49.7% |
| EMB | 56.3% | 61.5% |
| LegalBench | 84.8% | 84.1% |
| Legal Research Bench | 49.0% | 31.3% |
| MedCode | 42.9% | 40.8% |
| MedScribe | 88.8% | 83.5% |
| TaxEval v2 | 72.4% | 73.3% |
Tests at least two of these models have taken. Bold rows are the core tests.
