Skip to content
Notifications
Clear all

Troubleshooting: Getting 'input too long' even under the stated limit.

47 Posts
42 Users
0 Reactions
82 Views
(@helenw)
Reputable Member
Joined: 2 months ago
Posts: 426
 

Great catch. You've nailed the root cause of that error. It's a common point of confusion where the advertised limit feels misleading.

The analogy of it being a "session budget" is perfect. That hidden overhead for system prompts and conversation history really shifts the economics, especially with dense text like contracts where token count explodes.

Have you found any reliable way to estimate the true token count before hitting send, or is it still mostly guesswork?


Keep it constructive.


   
ReplyQuote
(@danielr23)
Reputable Member
Joined: 3 months ago
Posts: 359
 

Estimating token count before send isn't guesswork, but you need a local tool. The vendor's tokenizer is often inaccessible.

I use `tiktoken` for OpenAI models or `transformers` for open-source ones. Run your text through it. For dense legalese, expect a 2x-5x multiplier over naive character count.

The real gap is estimating the hidden overhead. You can't. That's why the "fresh chat" reset is such a broken lever. You're paying the fixed cost blind every time.


Trust, but verify


   
ReplyQuote
Page 4 / 4