Aider: Model gpt-3.5-turbo has hit a token limit!

What it means

The API provider reported that the request went over the model's context window (input plus output) or its output token limit. Aider reports the error and suggests actions. The token counts it prints are estimates. Aider never enforces token limits; it only reports token limit errors from the API provider. Lines such as "-- exceeded output limit!" or "-- possibly exhausted context window!" show which limit was hit.

Causes

• Trying to send too much data to a model, overflowing its context window (official docs).
• Asking for a large change that affects a lot of code, so the LLM hits output token limits; most models have output limits as low as 4k tokens (official docs).
• Non-compliant API proxy servers or bugs in the API server hosting a local model (official docs).

How to fix it

1. 1
   Fix 1.

   Reduce input tokens: use /tokens to see usage, /drop to remove unneeded files, /clear to clear chat history, and break your code into smaller source files.

   Snippet

   Copy

   /tokens
   /drop
   /clear

2. 2
   Fix 2.

   Reduce output tokens: ask for smaller changes in each request, break code into smaller files, use a strong model like gpt-4o, sonnet or DeepSeek V3 that can return diffs, or use a model that supports infinite output.

3. 3
   Fix 3.

   Try aider without an API proxy server, or directly with one of the recommended cloud APIs. For local models the docs note that Ollama is known to work well with aider.