Up to four models
Claude Opus 4.8 vs GLM-5.3
On the index (GPT-4 = 100), Claude Opus 4.8 scores 170 and GLM-5.3 168. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
Claude Opus 4.8
170154–186
WeightsClosed
ReleasedMay 2026
Price in / out$5.0 / $25
Context1M
Core tests taken6 of 9
168152–184
WeightsOpen
ReleasedAug 2026
Price in / out$0.070 / $1.4
Context1.05M
Core tests taken5 of 9
Claude Opus 4.8 and GLM-5.3 are too close to call (2 points apart, within the uncertainty), ahead on 2 of 5 shared core tests, at 25× the price.
Test by testthird-party results
Claude Opus 4.8
GLM-5.3
Every result as a table
| Test | Claude Opus 4.8 | GLM-5.3 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 91.0% | 90.9% |
| LiveBench Data Analysis | 66.0% | 70.2% |
| LiveBench IF | 72.0% | 69.3% |
| LiveBench Language | 79.7% | 79.9% |
| LiveBench Reasoning | 89.2% | 85.8% |
| GPQA Diamond (Vals AI) | 92.4% | 88.1% |
| MMLU Pro | 89.6% | 86.8% |
| SimpleBench | 64.8% | 66.2% |
| SimpleQA Verified | 53.0% | 41.0% |
| Maths | ||
| FrontierMath (tiers 1-3) | 80.0% | 68.8% |
| AIME-style maths (OTIS mock) | 98.3% | 91.1% |
| LiveBench Mathematics | 94.3% | 87.9% |
| Chess puzzles | 52.0% | 37.0% |
| FrontierMath (tier 4) | 56.1% | 29.3% |
| Coding | ||
| Terminal-Bench 4.0 | 23.2% | 38.9% |
| LiveBench Agentic Coding | 50.5% | 60.9% |
| LiveBench Coding | 81.8% | 79.0% |
| Code Migration | 47.3% | 44.2% |
| LiveCodeBench | 87.8% | 80.5% |
| ProgramBench | 1.0% | 1.5% |
| SWE-bench | 88.6% | 95.4% |
| Vibe Code Bench v1.1 | 82.7% | 78.1% |
| DeepSWE | 59.0% | 69.0% |
| WeirdML | 82.9% | 75.4% |
| Agents | ||
| APEX-Agents | 48.9% | 56.6% |
| SkillsBench | 59.2% | 47.5% |
| Terminal-Bench-Science | 4.3% | 5.7% |
| DTBench | 94.9% | 87.7% |
| LMCA | 57.5% | 55.5% |
| Mystery game puzzles | 36.0% | 33.0% |
| Professional work | ||
| Finance Agent (v2) | 53.9% | 55.8% |
| EMB | 69.4% | 56.3% |
| LegalBench | 83.6% | 84.8% |
| Legal Research Bench | 43.8% | 49.0% |
| MedCode | 53.2% | 42.9% |
| MedScribe | 85.8% | 88.8% |
| Public Benefits Bench v1.1 | 68.1% | 68.5% |
| Tax Agent Bench | 30.0% | 43.3% |
| TaxEval v2 | 75.6% | 72.4% |
Tests at least two of these models have taken. Bold rows are the core tests.