February 2026
Introducing bear-1.1: Improved LLM Compression
bear-1.1 is an improved version of the bear-1 compression model. It delivers better accuracy preservation and faster compression speeds while maintaining the same simple API integration.
What's improved in bear-1.1?
bear-1.1 builds on everything that made bear-1 effective and improves on it. The core approach is the same — semantic compression that removes redundant tokens before they reach your LLM — but bear-1.1 does it better.
Better accuracy preservation
bear-1.1 is more precise about which tokens to remove. It better understands context dependencies across long inputs, resulting in higher accuracy on downstream LLM tasks compared to bear-1.
Faster compression speeds
Optimized inference pipeline reduces compression latency. You get the same level of token reduction with lower overhead added to your LLM pipeline.
Same simple API
bear-1.1 is a drop-in replacement for bear-1. Change the model parameter from bear-1 to bear-1.1 in your API call and you're done.
Performance
bear-1.1 maintains the same compression ratios as bear-1 while improving output quality. The key metrics:
66%
Token reduction
3x
Cost reduction
With bear-1.1, compressed prompts preserve more of the semantic signal that matters. This means your LLM sees cleaner input, leading to more accurate and relevant responses.
Upgrading from bear-1
If you're already using bear-1, upgrading to bear-1.1 is a one-line change. Update the model parameter in your compression API call:
- "model": "bear-1"
+ "model": "bear-1.1"
The request and response format is identical. No other changes needed. bear-1 remains available for existing integrations.
When to use bear-1.1
We recommend bear-1.1 for all new integrations. It's the better model — improved accuracy preservation, lower latency, and the same cost. Use bear-1.1 when you need:
- •Chat applications with long conversation histories
- •Document processing and RAG pipelines
- •Any LLM workflow where you want to reduce token costs
Get started with bear-1.1
Create an account and start compressing with the latest model.