Incorrect Context Window Size Detection for VS Code LM API Provider Models: Request to Consider Improvement

[OleynikAleksandr]

Dear Roo Code Team,

I would like to request that you consider correcting the context window size detection for models available via the VS Code LM API provider.

Currently, there appears to be a discrepancy: for example, the Gemini 2.5 Pro model has an actual context window size of 1 million tokens, whereas Roo Code indicates it as 63 800 tokens.

This difference can significantly limit the usability of the model through your tool.

I would be grateful if you could look into this issue and, if possible, make the necessary adjustments.

Thank you!