Personally I've found these bigger models (o3/Claude 4 Opus) to be disappointing for coding.
Opus is really great but through Claude Code. If you used Cursor or RooCode it could be normal that you get disappointed
This matches my experience, but cant explain it. Do you know what's going on?
My understanding is context size. Companies like Cursor are trying to minimize the amount of context sent to the models to keep their own costs down. Claude Code seems to send a lot more context with every request and that seems to make the difference.
Just guessing, but the new Opus was probably RL tuned to work better with Claude Code's tool calls
I got the opposite experience. Not with Opus (too expensive), but with Sonnet. I got things done way more efficiently when using Sonnet with Roo than with Claude Code.
same. i ran a few tests ($100 worth of api calls) with opus 4 and didn’t see any difference compared to sonnet 4 other than the price.
also no idea why he thinks roo is handicapped when claude code nerfs the thinking output and requires typing “think”/think hard/think harder/ultrathink just to expand the max thinking tokens.. which on ultrathink only sets it at 32k… when the max in roo is 51200 and it’s just a setting.
I think I could share a trick that could help:
From my experience (so not an ultimate truth) Claude is not so great at taking the decision for planning by its own: it dives immediately into coding.
If you ask it to think step-by-step it still doesn’t do it but Gemini 2.5 Pro is good at that planning but terrible at actual coding.
So you can use Gemini as planner and Claude as programmer and you get something decent on RooCode.
This “think wisely” that you have to repeat 10x in the prompt is absolutely true
I think you misread my comment. I wasn't asking for help. I get consistent good output from Sonnet 4 using RooCode, without needing Gemini for planning.
Edit: I think I know where our miscommunication is happening...
The "think"/"ultrathink" series of magic words are a claudecode specific feature used to control the max thinking tokens in the request. For example, in claude code, saying "ultrathink" sets the max thinking tokens to 32k.
On other clients these keywords do nothing. In Roo, max thinking tokens is a setting. You can just set it to 32k, and then that's the same as saying "ultrathink" in every prompt in claudecode. But in Roo, I can also setup different settings profiles to use for each mode (with different max thinking token settings), configure the mode prompt, system prompt, etc. No magic keywords needed.. and you have full control over the request.
Claude Code doesn't expose that level of control.
i found them all disappointing in their own ways. Atleast deepseek models actually listen to what i say instead of ignoring me doing their own thing like a toddler.