Opinions: Claude Sonnet VS Opus

[bhaumikmaan]

Following #53 and discussion b/w code owners @dev-opsss & @totargaming
Let's discuss which of the 2 models we should have UNLESS we want both and can implement it flawlessly.

My 2 cents are

• We don't use -latest at all because we want to use stable versions

Now the question remains whether or not to use Claude opus or upgrade to sonnet

Feature	Claude 3.7 Sonnet (Feb 2025)	Claude 3 Opus (Feb 2024)
Price	Lower	Higher
Speed	Faster	Slower
Expertise 1	Good for quick solutions	Good for deep step-by-step solutions
Expertise 2	Good at volume fulfillment	Improved at catching edge cases

How I see it, is as a tradeoff of speed and price to correct solutions
Where I would prefer to choose correct solutions + opus is safe and reliable and tested.
Although Sonnet has advantages, we can add it as a choice or something, but it would make the code a bit shabby.
Opinions welcome

[totargaming]

I agree with this, would having these two together be the best idea? or only one of these.

[bhaumikmaan]

I agree with this, would having these two together be the best idea? or only one of these.

See, there's 2 views to this:

1. Do you want to run with the working product? Answer: Only one of those
2. Do you want to try your hand out and get the real feel of open source? Answer: Both

So your call to implement

How I would like to ship the PR is that we decide on a version i.e. Opus or Sonnet and close the PR there
Then open a new one for adding both if we want to do that

[dev-opsss]

We can for now ship with the sonnet 3.7 as its faster, Users always have an option to make the changes based on what they want. i tried with opus it works flawlessly, at the same time sonnet is also doing great

If we compare the token, Sonnet is the best one to understand the problem. If user is expecting complex problem they can use Opus for that. but having sonnet as primary will not cause any issues. it is not a bad model.

Token Limits Comparison

Claude 3.7 Sonnet:

• Context window: 200K tokens
• Max output:
  • Normal mode: 8192 tokens
  • Extended thinking mode: 64000 tokens
• Cost: $3.00 / $15.00 per million tokens (input/output)

Claude 3 Opus:

• Context window: 200K tokens
• Max output: 4096 tokens
• Cost: $15.00 / $75.00 per million tokens (input/output)

[bhaumikmaan]

Cool, sounds good to me