muse wonder vs swe 2 tested both models with the same prompt at highest reasoning available but came out really different > muse took 24 mins and cost about $18 > swe took 15 mins and cost zero dollars really surprised by how good swe 2 did here but... i don't know what muse wonder was doing model is scored above astra and fable 5.1 but it’s performing way worse than both in testsshow more

this is muse wonder full on max reasoning. the one that ranked higher then astra and fable 5.1 in benchmarks prompt was: create a beautiful, busy, living koi pond using Three.js in a single HTML file. Include a waterfall, rocks, plants, and multiple koi fish with different colors and patterns. Make the scene feel alive through continuous fish movement, flowing water, and animated environmental details. Add a “Feed the koi” button. When clicked, food should appear on the water, and the fish should visibly swim toward it and eat it—not simply continue their normal swimming animation. Allow the camera to orbit and zoom so I can explore the pond. Make the scene visually polished, with convincing water, lighting, and depth. Deliver the complete working HTML file.

bro wtf is that 😭😭

you wanna see more? all are coming out way worse no idea why. wtf is going on

I will share you few prompts...

i am not testing it's blocked here so i am having people try lol

ohh okay

benchmark shows reality

benchmarks 🫣

benchmaxxed 😭😭

first time hearing of swe - company behind them?

devin @cognition

Swe wins easily

This is Muse wonder yeah not the Full variant?

full on max

Whatt?? But how i expected more tbh

Sooo Devin sub + Claude sub still the meta right now 😄 And I guess opencode go if you want glm 5.3 or flash for more cyber focused work

Is this muse wonder full or max?

muse wonder full in muse code at max reasoning

lmao you just drop the video

That’s because benchmarks are fake @Ananth7e

swe is really cool and so cost efficient

Ive seen other tests and muse wonder seems not to be good.

clearly this isn't a competitor for astra, fable, and even argon just by looking at those tests

They all tune models for benchmarks; it's like they think people are stupid and won't test them on real-world tasks.

Slower and $18 vs faster and free is a real tradeoff. Curious if the quality gap justified the bill on your task.

Please share prompt jazii I want to add to my website

Create a beautiful, busy, living koi pond using Three.js in a single HTML file. Include a waterfall, rocks, plants, and multiple koi fish with different colors and patterns. Make the scene feel alive through continuous fish movement, flowing water, and animated environmental details. Add a “Feed the koi” button. When clicked, food should appear on the water, and the fish should visibly swim toward it and eat it—not simply continue their normal swimming animation. Allow the camera to orbit and zoom so I can explore the pond. Make the scene visually polished, with convincing water, lighting, and depth. Deliver the complete working HTML file.

Thank youu

I hope muse wonder isn't benchmaxed

Ooff

what stack is this?

Oh dear don’t tell me…BenchMaxxMasterFlex3000

I think swe cooking well bz cost good output

The useful comparison isn't just quality — it's $/task. If SWE 2 matched Muse on your prompt at $0 vs ~$18, that's the new default for anything that isn't Muse-specific. Run a 5-prompt bakeoff on your real workload and keep the cheaper one that scores within ~10%.

Thank you for doing this

muse giving this pixel feeling