Up to four models
Qwen3.8-27B vs GLM-5.2
On the index (GPT-4 = 100), Qwen3.8-27B scores 155 and GLM-5.2 154. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
155139–171
WeightsOpen
ReleasedAug 2026
Price in / out$0.060 / $2.2
Context1.02M
Core tests taken3 of 9
154138–169
WeightsOpen
ReleasedJun 2026
Price in / out$0.0070 / $6.8
Context1.05M
Core tests taken6 of 9
Qwen3.8-27B and GLM-5.2 are too close to call (2 points apart, within the uncertainty), ahead on 2 of 3 shared core tests, and costs 2.9× less.
Test by testthird-party results
Every result as a table
| Test | Qwen3.8-27B | GLM-5.2 |
|---|---|---|
| Reasoning & knowledge | ||
| LiveBench Data Analysis | 76.6% | 73.7% |
| LiveBench IF | 72.7% | 62.3% |
| LiveBench Language | 74.4% | 76.2% |
| LiveBench Reasoning | 80.0% | 78.6% |
| GPQA Diamond (Vals AI) | 88.9% | 85.6% |
| MMLU Pro | 84.3% | 86.7% |
| SimpleBench | 60.2% | 58.8% |
| Maths | ||
| LiveBench Mathematics | 86.2% | 89.8% |
| Surface Evolver Bench | 45.0% | 55.6% |
| Coding | ||
| LiveBench Agentic Coding | 61.4% | 51.8% |
| LiveBench Coding | 75.7% | 79.7% |
| Code Migration | 14.2% | 37.9% |
| LiveCodeBench | 84.0% | 69.5% |
| ProgramBench | 0.0% | 0.5% |
| SWE-bench | 86.0% | 82.8% |
| Vibe Code Bench v1.1 | 64.8% | 64.0% |
| Agents | ||
| APEX-Agents | 47.5% | 45.2% |
| SkillsBench | 38.1% | 45.1% |
| DTBench | 88.0% | 93.6% |
| LMCA | 41.4% | 45.8% |
| Professional work | ||
| Finance Agent (v2) | 48.5% | 49.7% |
| EMB | 59.7% | 61.5% |
| LegalBench | 82.4% | 84.1% |
| Legal Research Bench | 36.1% | 31.3% |
| MedCode | 28.7% | 40.8% |
| MedScribe | 83.9% | 83.5% |
| TaxEval v2 | 70.9% | 73.3% |
| Novel problems | ||
| ARC-AGI-2 | 42.4% | 22.8% |
| ARC-AGI-1 | 87.5% | 77.0% |
Tests at least two of these models have taken. Bold rows are the core tests.