/think-tokens
Sets the model's thinking-token budget for extended reasoning.
/think-tokens <value> sets how many tokens a model with extended thinking (e.g. Claude models with thinking support) may use for its internal reasoning process. Values can be given compactly, e.g. 8096, 8k, 10.5k, or 0.5M; 0 disables thinking. On the command line this corresponds to --thinking-tokens VALUE.
✅ WHEN TO USE IT?
Give more thinking budget for an architecturally complex question
/think-tokens 16k before a question about redesigning the data model layer
Turn thinking off specifically to get faster/cheaper answers
/think-tokens 0 for simple, repetitive changes across many small files
⛔ WHEN NOT TO?
The active model doesn't support extended thinking
The parameter only has an effect on models with a matching thinking feature, otherwise it does nothing.
Better: Use /reasoning-effort for models that instead support reasoning_effort levels (e.g. OpenAI's o-/GPT-5 series)
Setting a very high thinking budget by default for every request
An unnecessarily high budget increases latency and cost even when the task is trivial.
Better: Only raise the budget for genuinely demanding requests, then lower it again
SOURCES
- Aider Docs — In-chat Commands ↗ aider.chat