Optimize your AI costs with intelligent token compression. Our proven algorithms reduce usage while maintaining quality.
Start optimizing your AI costs today
We trim down the text you send to LLMs without messing up the quality. Less text means lower costs and helps your LLM function - it's that simple!
Our algorithms can get you nearly the same results with far fewer input tokens. Here's how we do the magic:
Fewer input tokens means your AI bills shrink dramatically. Use cases with a large context window could see a 80% reduction in costs.
We ensure your minimised prompt maintains at least 90% of the original meaning and intent.
Just call our API with a prompt, we'll optimize it and give it back. If we can't optimize, it won't cost you.
Our optimization algorithms have been rigorously tested on industry-standard benchmarks, demonstrating significant token reduction while maintaining high accuracy.
Boolean Questions
Recognizing Textual Entailment
CommitmentBank
Choice of Plausible Alternatives
Words in Context
Stop paying per input token. Only pay when you save.
Test our APIs and see how much you can save.
This is 80% cheaper than using ChatGPT directly.
We count tokens based on the standard GPT tokenization method with tiktoken. On average, one token is approximately 4 characters or 0.75 words in English.
When tuning our algorithm, we use the industry standard benchmark SuperGlue. We compare the performance of optimized prompts against the original. We find that the optimized prompt is nearly as performant and sometimes more performant. For individual projects, we measure the similarity between optimized prompts and the original by comparing the embeddings of the prompts.
Some uses cases, like writing code or precise prompt engineering, aren't a good fit for optimizing. We regularly run automated checks on the responses that we give to each user, to ensure that the same performance is maintained. If we detect a degradation, we pause optimizations and let you know.
Using TokenCrush can save significant amounts of money. However, you may see a performance degradation in some cases. See the compabability quiz to understand if TokenCrush is right for you.
Take our quick compatibility quiz to find out if your LLM setup works with TokenCrush
Natural language refers to human-readable text like questions, instructions, or content for the LLM to process.
Start reducing your AI costs with intelligent token optimization