Gemini 4 Carbon
Long-horizon work. Deep engineering, mathematical and scientific reasoning.
Which reasoning effort setting?
controls how long Gemini 4 Carbon thinks before answering. Token prices never change; higher settings spend more and raise task cost.
Start at Medium. It scores 58.3 against 60 at the top setting, for 15% less per benchmark task. Rule: cheapest setting within 2 points of the model’s best score.
At Medium, Carbon thinks for about 56.2K tokens per task, then writes about 46.8K more as its answer: $6.59 per task, 0.9× Low, for +4.1 points.
Output rate: $10.00/1M at every effort level. Thinking tokens are billed as output.
| Effort | Index | Per task | Thinking | All output | × Low | Time |
|---|---|---|---|---|---|---|
| Low | 54.2 | $7.19 | 32.4K | 86.2K | 1.00× | 28.4 min |
| MediumStart here | 58.3 | $6.59 | 56.2K | 103K | 0.92× | 28 min |
| High | 60.0 | $7.76 | 72.2K | 123K | 1.08× | 37.3 min |
Per task of the Artificial Analysis Intelligence Index: the average over its nine evaluations, counting input, cache, and output tokens at list prices. These are long, agent-style benchmark tasks, much longer than a chat message. Artificial Analysis prices this model at its 1 December rate ($4 / $20), so at today’s introductory price each task costs half.
Where Gemini 4 Carbon sits on the frontier
Intelligence Index against one 100K-input, 10K-output request.