I opened the Aistudio welcome page today; it shows the “2M token context window” on there.
But only the Gemini 1.5 Pro can provide this function, not the newer models, which is disappointing. 
Will newer 2M token context window model be released in the future?
Thank you for flagging this. Could you please let us know about your use case and what context length you are looking for?
This will help us better understand your needs and provide more relevant suggestions or feedback to the internal team.
You know, 2 million is not bad. However, now I have started to notice that gemini 2.5 pro 05-06 has become less stable in responses. I don’t know how it is in English, but it occurs in Russian.:
Thank you for your follow-up and for asking specifically about the context length I’m looking for.
To directly answer your question: given the current impact of Chain-of-Thought (CoT) and the model’s general verbosity, I believe restoring the originally highlighted 2 million token context window is the most immediate and crucial step.
While I haven’t deeply analyzed if even more than 2M would be beneficial for even more ambitious tasks (though intuitively, it seems plausible), my current focus and request is to at least have the 2M token capacity available. Here’s why this is so important based on my experience:
Impact of Chain-of-Thought (CoT) and Verbosity:
This is a critical factor. I’ve observed that the Gemini 2.5 Pro and Flash versions, often by default, include detailed Chain-of-Thought reasoning in their responses. While CoT is valuable for transparency, it significantly and rapidly consumes the context window. Even with a 1M window, this verbosity and the inclusion of CoT means the actual usable context for conversation history or complex instructions becomes very limited, very quickly.
Therefore, having the 2M token window would, at a minimum, help to mitigate this effect and provide a more practical context depth for the kind of applications I outlined below. Without it, even moderately complex interactions can quickly exhaust the 1M window due to the model’s own output style.
Long-term Conversational Partner & Life Guidance:
To maintain a coherent and deeply contextualized relationship with the bot over extended periods (months or years), even with just a few exchanges daily. A longer window would be essential for the bot to retain a rich understanding of our cumulative interactions.
Mastering New Skills (e.g., a new programming language):
This is a very important use case. A longer context window is vital for the model to effectively track learning progress, remember previously understood concepts versus those still needing work, and recall past examples or teaching approaches. This avoids repetitive instruction and makes the learning process far more efficient. The current CoT verbosity makes this particularly challenging with only 1M tokens, as much of the context is taken up by the model’s reasoning rather than the learning material itself.
In summary, the primary reason for requesting the 2M context window isn’t just about aspiring to larger theoretical limits, but about addressing a very practical current constraint: the rapid token consumption due to features like Chain-of-Thought. Restoring the 2M context window feels like a necessary step to achieve the level of utility that was implied and that is needed for these more advanced use cases.
I hope this clarifies my needs.
Thank you for your detailed feedback and for outlining the challenges you’re facing with the current 1M token context window. We appreciate your input and will share your concerns with the internal team for further review.
