Incorrect amount of reserved tokens for output on OpenRouter models

[daniel-lxs]

App Version

3.17.0

API Provider

Anthropic

Model Used

Any OpenRouter model that has the same output tokens as the context window

🔁 Steps to Reproduce

1. Select Deepseek v3 0324 from OpenRouter
2. Select the GMICloud provider or any provider that serves the model with the same amount of output tokens as the context window.
3. The reserved tokens will equal to the tokens of the context window leaving 0 available tokens.

💥 Outcome Summary (Optional)

The reserved tokens should be a fraction of the total context window.

📄 Additional Context

I addressed this issue on #3430 by adding a condition where if the maxToken is equal to safeContextWindow then the reserved tokens should be 20% of the total context window.

Commit: 3296bd6

CC

@cte

[dosubot]

The issue you're experiencing with the reserved tokens being equal to the context window size in OpenRouter models might be related to the configuration or logic that calculates token reservation. Unfortunately, I couldn't find specific details in the Roo Code repository regarding the implementation of token reservation logic or its configuration.

However, the custom system prompt override feature might be influencing the token reservation process. This feature allows dynamic prompts using context variables, and it is configured in the Roo Code UI under the "Advanced: Override System Prompt" section ^[1][2]. You might want to check if the override is affecting the token calculation.

If the issue persists, consider checking for any open issues or pull requests in the repository that might address this problem, or submit a bug report for further investigation.

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Bug Report | Other}