Thinking levels · xAI
Does Grok 4 Fast write better when it thinks harder?
It does, and by a lot for a model with Fast in its name. Grok 4 Fast (reasoning) beats the plain Grok 4 Fast by 182 Elo. All it asks for is $0.029 instead of $0.022 per script, plus a bit more waiting.
Both settings we ran
| Setting | Elo | Step | Overall | Cost / script | Time / script | Reasoning tokens | Board rank |
|---|---|---|---|---|---|---|---|
| high | 1011 ±50 | - | 73.6 | $0.029 | 0.9 min | 3,272 | #145 |
| default (no effort flag) | 830 ±44 | - | 70.2 | $0.022 | 0.4 min | 947 | #154 |
Which setting should you use?
Use Grok 4 Fast (reasoning). The two Elo ranges don't even touch, so that's strong evidence the reasoning helps, and the extra cost is under a cent per script. The plain version earns its name at 0.4 minutes, but a quick weak draft still costs you the time to fix it, and it lands at #154 on the board. To be honest, even the reasoning version only reaches #145, so I'd treat both as cheap baselines to compare against, not models I'd ship scripts with.
Other xAI ladders
Elo comes from comparing every pair of models on every script across the full board, so these settings sit on the same scale as the leaderboard. Cost is one full script at API list prices, and time is one accepted script, retries included. Compare this ladder with other models on the thinking-levels page, or see every xAI model side by side.