Speaking of, I think Vals underrates Qwen 3.7 Max. It's one of the strongest Chinese models overall, but pulled down by ridiculously low Vibe Code Bench v1.1. Like, it's below its lesser open source siblings. 3.7 *Plus* gets 46.4 there. What's up?
For those looking into open weight models in light of recent news … we’ve just evaluated Kimi K2.7 Code on the Vals coding benchmarks
Users are excited about Kimi's rapid iteration gains on the Vibe Code Bench and thank ValsAI for quickly fixing the related bug.
Glad to have that fixed Still, VCB 1.1 is interesting in that it's probably the only eval where DSV4-Pro is straightforwardly the best Chinese open model (and best Chinese OR open model). Feels about right but I'm biased ofc.
@teortaxesTex Thanks for catching this @teortaxesTex, this was indeed a bug on our site- note that this did not affect Vals Index, and we have fixed the score on VCB: https://www.vals.ai/models/alibaba_qwen3.7-max
@teortaxesTex Thanks for catching this @teortaxesTex, this was indeed a bug on our site- note that this did not affect Vals Index, and we have fixed the score on VCB: https://www.vals.ai/models/alibaba_qwen3.7-max
@ValsAI Thanks is Nemotron similar or is it actually that low?
@teortaxesTex It’s wild how much kimi gains on each iteration. An opus iteration barely did anything but a kimi iteration might as well be a new gen.
@teortaxesTex Vals post about the mimo models says that v2.5 ranks higher than 2.5 pro specifically on vibe code bench because 2.5 pro lacks vision
@teortaxesTex 确实如此