| Grok 4 Heavy | Grok 4 | |
|---|---|---|
| arenaElo | 1391 | 1341 |
| humanEval | 88.4 | 85.1 |
| sweBench | 55 | 50 |
| mmlu | 89.3 | 87.5 |
| gpqa | 74.5 | 57 |
| aime | 92 | 88 |
| Price in/out ($/M) | $15 / $60 | $3 / $15 |
Heavy gains ~17 GPQA points over vanilla Grok 4 by running debate among copies of itself. Price is ~5× higher. For research and benchmark-driven evals, Heavy. For standard chat and social agents, regular Grok 4.