Up to four models
GLM-5.3-Flash vs Grok 4.5
On the index (GPT-4 = 100), GLM-5.3-Flash scores 161 and Grok 4.5 161. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
161145–177
WeightsOpen
ReleasedAug 2026
Price in / out$0.15 / $0.50
Context1.05M
Core tests taken6 of 9
Grok 4.5
161145–177
WeightsClosed
ReleasedJul 2026
Price in / out$2.0 / $6.0
Context500K
Core tests taken6 of 9
GLM-5.3-Flash and Grok 4.5 are too close to call (0 points apart, within the uncertainty), ahead on 3 of 6 shared core tests, and costs 13× less.
Test by testthird-party results
Every result as a table
| Test | GLM-5.3-Flash | Grok 4.5 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 90.1% | 93.4% |
| LiveBench Data Analysis | 76.4% | 73.0% |
| LiveBench IF | 52.8% | 71.5% |
| LiveBench Language | 77.3% | 82.8% |
| LiveBench Reasoning | 77.6% | 87.2% |
| GPQA Diamond (Vals AI) | 86.4% | 92.9% |
| MMLU Pro | 86.1% | 89.2% |
| MMMU Pro | 86.0% | 61.8% |
| Maths | ||
| FrontierMath (tiers 1-3) | 55.8% | 57.2% |
| AIME-style maths (OTIS mock) | 93.9% | 97.8% |
| LiveBench Mathematics | 81.2% | 90.8% |
| ProofBench v1.1 | 21.0% | 31.0% |
| Chess puzzles | 21.0% | 35.0% |
| FrontierMath (tier 4) | 17.1% | 24.4% |
| Surface Evolver Bench | 52.5% | 74.4% |
| Coding | ||
| Terminal-Bench 4.0 | 37.4% | 8.6% |
| LiveBench Agentic Coding | 56.8% | 56.5% |
| LiveBench Coding | 79.0% | 68.6% |
| Code Migration | 20.5% | 36.6% |
| CyberBench v1.1 | 67.9% | 71.0% |
| IOI | 52.5% | 40.6% |
| LiveCodeBench | 80.5% | 87.4% |
| ProgramBench | 0.0% | 0.0% |
| SWE-bench | 92.0% | 86.6% |
| Vibe Code Bench v1.1 | 30.8% | 69.0% |
| DeepSWE | 63.4% | 53.8% |
| Agents | ||
| APEX-Agents | 52.8% | 56.2% |
| SkillsBench | 40.2% | 66.0% |
| SRE Bench | 0.8% | 0.8% |
| Professional work | ||
| Finance Agent (v2) | 57.9% | 48.4% |
| EMB | 55.9% | 52.9% |
| LegalBench | 83.9% | 86.0% |
| Legal Research Bench | 45.2% | 38.0% |
| MedScribe | 88.9% | 86.9% |
| Tax Agent Bench | 31.6% | 24.9% |
| TaxEval v2 | 75.6% | 71.7% |
| Novel problems | ||
| ARC-AGI-2 | 65.8% | 52.6% |
| ARC-AGI-1 | 91.0% | 87.2% |
Tests at least two of these models have taken. Bold rows are the core tests.