Thinking mode: when to turn it on
With thinking on, the model works through the task first and shows its reasoning above the answer. Reasoning tokens are billed like answer tokens, so use it for hard tasks only.
When it helps
Math, tricky code, multi-step logic and document review. For translations, emails or definitions it adds cost without improving quality.
Answer prices
Without thinking: a 1,000-token question and a 1,000-token answer. USD at the best rate: $200 = 10,000,000 tokens. Larger plans make tokens cheaper.
| Model | Tokens | USD |
|---|---|---|
| DeepSeek V4.1 Flash | 57 tokens | from $0.001 |
| GPT-5.6 Terra | 1,064 tokens | from $0.021 |
| Claude Sonnet 5.5 | 912 tokens | from $0.018 |
| Claude Opus 5.5 | 1,824 tokens | from $0.036 |