SONNET 5 TEST: A good AI Allrounder for cheap by Anthropic?

Keep Calm, and carry on burning those tokens

It's really annoying. It refuses perfectly normal things and twists things around. I've also noticed that it wastes a lot of tokens just trying to figure out whether what I'm doing is allowed or not. That's really bad

you get what you pay for - nothing in this case

I still much prefer Opus 4.6 because it understands my prompts so much better. In my experience Opus 4.6 is: much wiser with cross domain long questions, seems much more consistent with good replies, not as lazy or error prone, & frankly not as irritating & arrogant as 4.7, 4.8, and Sonnet 5. However, I don't use it for coding. So maybe those newer models are better suited for that.

That benchmark chart is getting ragged online because there was an earlier chart showing much less impressive results.

Based on my personal experience, I can say it's garbage. I can't believe I actually prefer the previous model. It lacks creativity, and it refuses to do simple corrections. If it misunderstands a request and writes the wrong text, then you point out the mistake, it often refuses to fix it, claiming it won't "change the truth" This is the worst AI I've used, and I'm genuinely disappointed

I think you guys all mean to start realizing that open AI acknowledging on their new model card, that they are having real problems with their model. This is not a fluke. This is not... something that is sudden. And now, Sonett is starting to show its side.

Only if you choose max thinking So many effort and thinking limits that it doesn’t have to cost more than opus

Can you test the deepseek 4 dspark release? They are claiming a major breakthrough.

I think usually sonnet is weaker than the previous opus. So that part is not different. What I don’t understand at all is who they want to charge more for sonnet than opus when opus is better.