Claude 3.7 Sonnet Support

[mwolson]

Claude 3.7 Sonnet just launched: https://www.anthropic.com/news/claude-3-7-sonnet

I wrote this code snippet to start using it early:

    (require 'gptel-anthropic)
    (unless (alist-get 'claude-3-7-sonnet-20250219 gptel--anthropic-models)
      (add-to-list 'gptel--anthropic-models
                   '(claude-3-7-sonnet-20250219
                     :description "Highest level of intelligence and capability" :capabilities
                     (media tool-use cache)
                     :mime-types
                     ("image/jpeg" "image/png" "image/gif" "image/webp" "application/pdf")
                     :context-window 200 :input-cost 3 :output-cost 15 :cutoff-date "2024-11")))

The knowledge cutoff date is taken from: https://support.anthropic.com/en/articles/8114494-how-up-to-date-is-claude-s-training-data

[lispy-ai]

And also support for Thinking Mode: #663

[endgame]

Already done? 2bb081e

[wlauppe]

Already done? 2bb081e

Indeed, this is already implemented. Kudos karthik! What a speed!
I upgraded, to the latest commit:
Claude 3.7 works flawlessly. Great 👍

What we should think about is how the new features can be expossed. Claude 3.7 has "extended thinking" available
there are two parameters:

extended_thinking_enabled: True/False
extended_thinking_budged_tokens: Integer

the budget_tokens should be the tokens used for thinking and always set less than the max_tokens.
in the api the vars are called slightly different, but i think it might sense to rename them if you expose them in emacs. The above is my suggestion. The final curl should look like this.

 "model": "claude-3-7-sonnet-20250219",
    "max_tokens": 20000,
    "thinking": {
        "type": "enabled",
        "budget_tokens": 16000
    },

Here are the complete docs.
https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking

Karthik, can you make this two parameters in the gptel-menu and general gptel api availabe. That would be great!

[endgame]

I would follow #663, which seems to be the discussion for "thinking mode".

[karthink]

I've added some instructions for using thinking mode to the README. As mentioned in #663, I'll add proper support to the UI and backend parsers along with the other APIs that provide reasoning models, but it will take a few weeks.

[karthink]

I've added support for including claude-3-7-sonnet-20250219's thinking/reasoning content with the response, or redirecting it elsewhere.

The user option is gptel-include-reasoning, and it should work for all backends/models that include reasoning blocks. As with most options you can set it from the transient menu with whatever scope you need.

Please let me know if it works as expected.

[karthink]

Closing as there are no additional actions planned for claude-3-7 right now.