Request to Increase Gemini CLI Context Limit Beyond 1M Tokens

[open54261]

Hello Gemini Team,

I hope you’re doing well.

I’m writing to raise a concern regarding the current context length limitation in Gemini, specifically the 1M token context. This same limit already existed in Gemini 2.5, and it has become a significant bottleneck—especially when using the Gemini CLI.

In practical development workflows, program files and repositories can become fairly large, and once they exceed a certain size, the Gemini CLI struggles or becomes unreliable. This severely limits its usefulness for real-world projects where full context across multiple files is essential.

Given the progress in model capabilities and the increasing complexity of modern codebases, I strongly believe that increasing the context window beyond 1M tokens would dramatically improve the developer experience and make Gemini far more competitive and practical for large-scale work.

Could someone from the Gemini support or engineering team please advise:

Whether there are plans to increase the context length beyond 1M tokens

If there are recommended workarounds or upcoming improvements specifically for handling large projects in the Gemini CLI.
I would also to add that the problem still occurs in gemini 3 too.

Thank you for your time and for the continued work on Gemini. I’d really appreciate any guidance or updates on this matter.

[Fohlen]

@open54261 in a transformer architecture the cost of doing computations is O(n^2) in regards to the input length. In other words, for every logarithmic increase in input, the amount of compute has to double. Since Gemini is a transformer model, don't expect any increment in context length in the foreseeable future. It is frustrating but it is a fundamental limitation of the technology. Try to process your software piecewise, wherever you can

[SkybuckFlying]

I recommend you try out the anti-gravity ide, it's agent seems to be able to dive into large code bases quite easily while still using models with relatively small context windows like 1m tokens ;)