Up to four models
Grok 4.20 vs Grok 4.3
On the index (GPT-4 = 100), Grok 4.20 scores 156 and Grok 4.3 135. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
Grok 4.20
156140–172
WeightsClosed
ReleasedFeb 2026
Price in / out$1.25 / $2.5
Context2M
Core tests taken4 of 9
Grok 4.3
135115–156
WeightsClosed
ReleasedApr 2026
Price in / out$1.25 / $2.5
Context1M
Core tests taken1 of 9
Grok 4.20 leads Grok 4.3 by 21 points, ahead on 0 of 1 shared core tests, at a similar price.
Test by testthird-party results
Grok 4.20Grok 4.3
Every result as a table
| Test | Grok 4.20 | Grok 4.3 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Vals AI) | 88.6% | 91.4% |
| MMLU Pro | 86.3% | 85.8% |
| MMMU Pro | 83.5% | 83.1% |
| Coding | ||
| Code Migration | 0.3% | 6.8% |
| LiveCodeBench | 84.3% | 84.5% |
| SWE-bench | 72.2% | 71.4% |
| Vibe Code Bench v1.1 | 4.1% | 19.4% |
| Professional work | ||
| Finance Agent (v2) | 28.5% | 37.7% |
| CorpFin v2 | 63.7% | 68.5% |
| EMB | 11.7% | 18.4% |
| LegalBench | 77.7% | 84.5% |
| Legal Research Bench | 13.9% | 15.4% |
| MedCode | 32.2% | 38.1% |
| MedScribe | 63.4% | 74.4% |
| MortgageTax | 45.4% | 48.3% |
| Tax Agent Bench | 7.0% | 6.2% |
| TaxEval v2 | 74.1% | 70.8% |
Tests at least two of these models have taken. Bold rows are the core tests.