Improved: Added support for reasoning models on OpenRouter
Improved: Better cache breakpoint handling for Anthropic models
Other bug fixes and improvements
What's new?
The most notable change in this release is supporting Anthropic's new Claude 3.7 model. This new model allows you to set a custom "thinking budget tokens". You can decide how hard the model think.
BoltAI simplifies this by using the same Reasoning Effort setting:
Low: 4K thinking tokens
Medium: 16k thinking tokens
High: 32k thinking tokens
It's recommended to start with medium (16k tokens) as it's the optimal setting when using Claude 3.7 extended thinking.
