It sounds very expensive on their side, feeding the whole thinking process from a cheaper model (R1) to a more expensive model (Claude) for example, would require a different type of billing or count as a double prompt.
Am I wrong?
Teal Lightbulb
It would be great to be able to select a reasoning model, and an output model. That way you can select a model like R1 for reasoning, and the hook it into sonnet 3.5 for output.
Google also have a really good and fast model for reasoning: gemini-2.0-flash-thinking-exp-01-21. It would be great to be able to combine this with other models.
This request was merged into another request
Comments3
Purple Teacup
It sounds very expensive on their side, feeding the whole thinking process from a cheaper model (R1) to a more expensive model (Claude) for example, would require a different type of billing or count as a double prompt.
Am I wrong?
Teal Lightbulb
It would be great to be able to select a reasoning model, and an output model.
That way you can select a model like R1 for reasoning, and the hook it into sonnet 3.5 for output.
Google also have a really good and fast model for reasoning: gemini-2.0-flash-thinking-exp-01-21. It would be great to be able to combine this with other models.
Reddit post talking about it too:
https://www.reddit.com/r/singularity/comments/1iec2p9/deepclaude_combines_claude_sonnet_35_with/
Purple Wombat
this should be relevant here: https://github.com/getAsterisk/deepclaude