Up to four models
GLM-5.3-Flash vs Grok 4.20
On the index (GPT-4 = 100), GLM-5.3-Flash scores 161 and Grok 4.20 156. 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 4.20
156140–172
WeightsClosed
ReleasedFeb 2026
Price in / out$1.25 / $2.5
Context2M
Core tests taken4 of 9
GLM-5.3-Flash leads Grok 4.20 by 5 points, ahead on 4 of 4 shared core tests, and costs 6.6× less.
Test by testthird-party results
Every result as a table
| Test | GLM-5.3-Flash | Grok 4.20 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 90.1% | 89.3% |
| GPQA Diamond (Vals AI) | 86.4% | 88.6% |
| MMLU Pro | 86.1% | 86.3% |
| MMMU Pro | 86.0% | 83.5% |
| Maths | ||
| FrontierMath (tiers 1-3) | 55.8% | 44.9% |
| AIME-style maths (OTIS mock) | 93.9% | 92.2% |
| Chess puzzles | 21.0% | 29.0% |
| FrontierMath (tier 4) | 17.1% | 17.1% |
| Coding | ||
| Code Migration | 20.5% | 0.3% |
| LiveCodeBench | 80.5% | 84.3% |
| SWE-bench | 92.0% | 72.2% |
| Vibe Code Bench v1.1 | 30.8% | 4.1% |
| Professional work | ||
| Finance Agent (v2) | 57.9% | 28.5% |
| EMB | 55.9% | 11.7% |
| LegalBench | 83.9% | 77.7% |
| Legal Research Bench | 45.2% | 13.9% |
| MedScribe | 88.9% | 63.4% |
| Tax Agent Bench | 31.6% | 7.0% |
| TaxEval v2 | 75.6% | 74.1% |
| Novel problems | ||
| ARC-AGI-2 | 65.8% | 65.1% |
| ARC-AGI-1 | 91.0% | 89.5% |
Tests at least two of these models have taken. Bold rows are the core tests.