News for hackers. Stuff that matters.

A comment on Qwen3-235B-A22B-Thinking-2507.

Re:Qwen3-235B-A22B-Thinking-2507 by sophia01 on Friday July 25, 2025 @08:55AM (#44682601)
If this is actually competitive with Gemini 2.5 Pro that would be insane esp for an Apache2 truly open weights model, let's hope it's not too hacked to shine on benchmarks!
[ Parent | on HN ]
Re:Qwen3-235B-A22B-Thinking-2507 by lvl155 on Friday July 25, 2025 @09:21AM (#44682852)
Qwen3 models are solid and at such a low cost, it doesn’t hurt to pair it with something like Sonnet 4 as a check. I mean it does eliminate a lot of Claude’s “You’re absolutely right!” loops.
[ Parent | on HN ]
Re:Qwen3-235B-A22B-Thinking-2507 by apwell23 on Friday July 25, 2025 @09:39AM (#44683003)
> I mean it does eliminate a lot of Claude’s “You’re absolutely right!” loops.

not as scary as "Let me try a completely different approach" . Now you have to throw out all the AI slop and start from scratch.

[ Parent | on HN ]
Re:Qwen3-235B-A22B-Thinking-2507 by cma on Friday July 25, 2025 @10:25AM (#44683483)
If you aren't using source control
[ Parent | on HN ]