Have a large context window, Chat switch to premium messages instead of blocking them all together. It’s pretty frustrating that I’m limited to 32K tokens for models that can take more.
Yeah, the truth is, if I can't upload a large document, the product instantly becomes useless to me. Because if I can't use it how I need to, even if it's only 10% of the time, I need to maintain a Claude subscription. So I simply can't subscribe to this product until it reaches feature parity with Claude and ChatGPT.
Chocolate Sand
Yes, I’d like to be able to to upload bigger code files for review by AI.
Mark Florkowski
Team
How much would you be willing to pay for this ability?
Teal Lightbulb
Im willing to pay a 25 dollars plan for unlimited context window
Amber Fuelcell
is it possible to make the file upload just cost some mana (base and overage)? I would really love to use the top models+reasoning with BYOK with the full supported context limits. Cuz upload conveniences will cost some for t3.chat. I’d happily keep the Pro sub and use both the sub and BYOK for prompts, as long as the Pro sub allows me to upload files and utilize the full context.
Mark Florkowski
Team
Will have to think on how we can allow this without it being super easy to use up large portions of your monthly limits accidentally. It’s very easy to attach a single file and make a request cost upwards of $1, which is simply untenable for an $8 subscription.
Amber Fuelcell
Makes sense. Separately, do you think there might be no “golden mean” between safety features and flexibility, for improving UX? So perhaps need a big scary red switch for the “i know what i’m doing” mode with such a danger disclaimer. Of course, it’s easy to just add settings, but then there are more things to keep track of for both sides.
Mark Florkowski
Team
Yes, I am trying to explore every posibility that we can to balance this, but it may just end up being a toggle in settings with a waiver.
Of course, it’s easy to just add settings, but then there are more things to keep track of for both sides.
The eternal struggle.
Magenta Peach
I hit that message: Your chat is too long. You can try shortening your last message or starting a new chat.
which seems to me to point to the limitations of the context window. Opening a new chat does not really help when coding, as the context is quite important.
So we need a method to:
shrink the context, like in the “Scrolling Context Window” proposal
select part of the current context to be taken with into the new chat
Of course some indication how large the current context is, how much is already used, would be very nice. yeah having a large context could move the problem a little bit out of sight ;-)
Mark Florkowski
Team
Have somethign to address this planned. Right now though I would recommend switching to a model like Gemini Flash (which has a very large context window) and either continuing with that model, or asking it for a summary which you can copy-paste into a new chat with your model of choice.
Sapphire Pinwheel
It is really ridiculous. Used the app often but now vonstantly getting the chat is to length error on 200k token uploads. Only a few models will accept them but no claude or gpt 5.5.
Mark Florkowski
Team
The limits are in place because requests like that would use plan limits extremely quickly on the $8 tier. Limits are increased on premium tier, and removed if you bring your own key.
Sapphire Pinwheel
I understand this but why is there no option to “override” this limit on my own behalf? I feel like i am competent enough to make my own decisions. I need this 4 to 5 times a month so wouldn't set me back a lot, cant say that for others though.
I have a subscription for chatGPT but would like to compare outputs of different models. Just a little toggle button? :)
Mark Florkowski
Team
cant say that for others though.
And this is why.
Just a little toggle button? :)
Maybe, though it’s a pretty niche use case. You can add an openrouter API key and set it to “fallback” mode, with this option in settings:
Log in to comment and vote
Comments16
Rose Exoplanet
Yeah, the truth is, if I can't upload a large document, the product instantly becomes useless to me. Because if I can't use it how I need to, even if it's only 10% of the time, I need to maintain a Claude subscription. So I simply can't subscribe to this product until it reaches feature parity with Claude and ChatGPT.
Chocolate Sand
Yes, I’d like to be able to to upload bigger code files for review by AI.
Mark Florkowski
How much would you be willing to pay for this ability?
Teal Lightbulb
Im willing to pay a 25 dollars plan for unlimited context window
Amber Fuelcell
is it possible to make the file upload just cost some mana (base and overage)? I would really love to use the top models+reasoning with BYOK with the full supported context limits. Cuz upload conveniences will cost some for t3.chat. I’d happily keep the Pro sub and use both the sub and BYOK for prompts, as long as the Pro sub allows me to upload files and utilize the full context.
Mark Florkowski
Will have to think on how we can allow this without it being super easy to use up large portions of your monthly limits accidentally. It’s very easy to attach a single file and make a request cost upwards of $1, which is simply untenable for an $8 subscription.
Amber Fuelcell
Makes sense. Separately, do you think there might be no “golden mean” between safety features and flexibility, for improving UX? So perhaps need a big scary red switch for the “i know what i’m doing” mode with such a danger disclaimer. Of course, it’s easy to just add settings, but then there are more things to keep track of for both sides.
Mark Florkowski
Yes, I am trying to explore every posibility that we can to balance this, but it may just end up being a toggle in settings with a waiver.
The eternal struggle.
Magenta Peach
I hit that message:
Your chat is too long. You can try shortening your last message or starting a new chat.
which seems to me to point to the limitations of the context window.
Opening a new chat does not really help when coding, as the context is quite important.
So we need a method to:
shrink the context, like in the “Scrolling Context Window” proposal
select part of the current context to be taken with into the new chat
Of course some indication how large the current context is, how much is already used, would be very nice.
yeah having a large context could move the problem a little bit out of sight ;-)
Mark Florkowski
Have somethign to address this planned. Right now though I would recommend switching to a model like Gemini Flash (which has a very large context window) and either continuing with that model, or asking it for a summary which you can copy-paste into a new chat with your model of choice.
Sapphire Pinwheel
It is really ridiculous. Used the app often but now vonstantly getting the chat is to length error on 200k token uploads. Only a few models will accept them but no claude or gpt 5.5.
Mark Florkowski
The limits are in place because requests like that would use plan limits extremely quickly on the $8 tier. Limits are increased on premium tier, and removed if you bring your own key.
Sapphire Pinwheel
I understand this but why is there no option to “override” this limit on my own behalf? I feel like i am competent enough to make my own decisions. I need this 4 to 5 times a month so wouldn't set me back a lot, cant say that for others though.
I have a subscription for chatGPT but would like to compare outputs of different models. Just a little toggle button? :)
Mark Florkowski
And this is why.
Maybe, though it’s a pretty niche use case. You can add an openrouter API key and set it to “fallback” mode, with this option in settings:
to achieve almost what you are looking for.