<![CDATA[A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard]]>This post did not contain any content.]]>https://citiverse.it/topic/efccae16-a3c6-4a77-a6d9-6628a181c9ae/a-team-of-5-qwen3.8-27b-models-beats-fable-5-on-livecodebench-hardRSS for NodeTue, 03 Nov 2026 06:31:06 GMTSun, 01 Nov 2026 16:05:40 GMT60<![CDATA[Reply to A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard on Mon, 02 Nov 2026 16:30:30 GMT]]>Money terms somewhere between 3k and 6k

]]>https://citiverse.it/post/https://lemmy.world/comment/26649072https://citiverse.it/post/https://lemmy.world/comment/26649072Mon, 02 Nov 2026 16:30:30 GMT<![CDATA[Reply to A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard on Mon, 02 Nov 2026 15:28:50 GMT]]>Yeah, that sound pretty usable, and seems like making the harness smarter can up the quality quite a bit. And completely agree about architecting things yourself. I find you want to basically do all the scaffolding, and lay out how the project looks like in terms of structure, the call graph, etc., and then let the model fill in the details. But also like you say, for something that's a throwaway, or a small personal project to scratch an itch, it doesn't really matter.

]]>https://citiverse.it/post/https://lemmy.ml/comment/28709361https://citiverse.it/post/https://lemmy.ml/comment/28709361Mon, 02 Nov 2026 15:28:50 GMT<![CDATA[Reply to A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard on Mon, 02 Nov 2026 15:23:52 GMT]]>I have nothing really to compare it to in terms of the big ones because i refuse to give anthropic/openai any money. But i have used this qwen model in opencode to port an old CUDA/Optix based project to HIP and that worked fairly well. Not an overly complex refactor to be honest because 99% is just replacing the cuda prefix everywhere with hip, but it did fix the CMakeLists.txt for me and ported the optix prime acceleration structure to hipRT as well. I also let it vibecode a react based portfolio page because i have absolutely 0 knowledge of web development and refuse to learn it because i think it is insane and that turned out perfectly fine as well.

So it is absolutely usable for real world projects and the performance is quite OK as well with ~35tok/s. If i up the context size to the full 256k window, the performance drops to 5 tok/s, which is ok, if i let it do a boring refactor over night but at that point it isn't faster than running DeepSeekV4.1-Flash-1017 through Strata and that one is much closer to the current SotA.

Definitely exciting times. We are close to being able to run very capable LLMs locally that are more than good enough for development. I am personally not a fan of 100% vibe coded projects, because i like to architect things myself, so letting it go in smaller steps and/or help be with the boring stuff is absolutely fine. Unless the project is some personal stuff that just needs to solve one simple problem than vibe coding is fine imho.

]]>https://citiverse.it/post/https://lemmy.ml/comment/28709264https://citiverse.it/post/https://lemmy.ml/comment/28709264Mon, 02 Nov 2026 15:23:52 GMT<![CDATA[Reply to A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard on Mon, 02 Nov 2026 15:06:23 GMT]]>Oh that's not bad at all. How do you find q4 performs overall?

]]>https://citiverse.it/post/https://lemmy.ml/comment/28709014https://citiverse.it/post/https://lemmy.ml/comment/28709014Mon, 02 Nov 2026 15:06:23 GMT<![CDATA[Reply to A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard on Mon, 02 Nov 2026 14:13:53 GMT]]>I've been running the Ridge-GGUF version on my RX 9070 XT with ~60-80k contextsize with q4 kv cache quantization. Everything fits completely in 16GB VRAM

]]>https://citiverse.it/post/https://lemmy.ml/comment/28708240https://citiverse.it/post/https://lemmy.ml/comment/28708240Mon, 02 Nov 2026 14:13:53 GMT<![CDATA[Reply to A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard on Mon, 02 Nov 2026 13:34:35 GMT]]>You'd need around 64gb vram to run 27b model at q6/8. You don't have to run multiple models at the same time though, since the same model can work all the different roles sequentially.

]]>https://citiverse.it/post/https://lemmy.ml/comment/28707671https://citiverse.it/post/https://lemmy.ml/comment/28707671Mon, 02 Nov 2026 13:34:35 GMT<![CDATA[Reply to A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard on Mon, 02 Nov 2026 13:28:56 GMT]]>What is the required hardware?

]]>https://citiverse.it/post/https://programming.dev/comment/26881846https://citiverse.it/post/https://programming.dev/comment/26881846Mon, 02 Nov 2026 13:28:56 GMT<![CDATA[Reply to A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard on Sun, 01 Nov 2026 21:01:12 GMT]]>I ended up rolling my own.

]]>https://citiverse.it/post/https://lemmy.ml/comment/28692034https://citiverse.it/post/https://lemmy.ml/comment/28692034Sun, 01 Nov 2026 21:01:12 GMT<![CDATA[Reply to A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard on Sun, 01 Nov 2026 17:23:41 GMT]]>Which harness you guys use with this qwen? I use it with open code 2, but it always end up in a loop after just a couple of prompts

]]>https://citiverse.it/post/https://feddit.org/comment/15732214https://citiverse.it/post/https://feddit.org/comment/15732214Sun, 01 Nov 2026 17:23:41 GMT