Hackernews posts about Qwen3

1. 01
   

   newRun Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/sgithub.com

   ▲ 923snehesht2d💬 421savesaved
2. 02
   

   newKev: Tiny Jev-like family of decision models built on top of Qwen3.5github.com

   ▲ 463tosh16d💬 210savesaved
3. 03
   

   newQwen 3.8 Omni Flashqwen.ai

   ▲ 346jjcm19d💬 138savesaved
4. 04
   

   newBenchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapsesquesma.com

   ▲ 286stared28d💬 139savesaved
5. 05
   

   newQwen 3.8 follows GPT-5.5 Pro reasoning prefillsgist.github.com

   ▲ 237wsxiaoys27d💬 93savesaved
6. 06
   

   newShapelearn Qwen 3.8 27B (13.1 GB VRAM)byteshape.com

   ▲ 104syntaxing19d💬 39savesaved
7. 07
   

   newShow HN: Nari Qwen3-TTS and Qwen3-ASR – High accuracy, low latency and costnarilabs.com

   ▲ 92toebee22d💬 31savesaved
8. 08
   

   newI made my iPhone a second GPU for my MacBook-Qwen 3.8 27B prefills 29–44% fasterreddit.com

   ▲ 39thunderbong3d💬 2savesaved
9. 09
   

   newShow HN: Swift-Qwen3.8-27B, -58.3% thinking, x1.95 speed, accuracy of xhighhuggingface.co

   ▲ 33kisjovan20d💬 17savesaved
10. 10
    

    newRun QWEN3.8 27B on 16gb Nvidia GPUsgithub.com

    ▲ 28Pragmata19d💬 10savesaved
11. 11
    

    newQwen3.8-Flash-Next non-uniform quantization runs on 2 RTX3090shuggingface.co

    ▲ 9pfeifferj22d💬 2savesaved
12. 12
    

    newGVS5H: Five Qwen3.8-27B Models Match Claude Fable 5 on LiveCodeBench Hardgithub.com

    ▲ 6OakNinja24ddiscusssavesaved
13. 13
    

    newQwen3.8 models outperforms Fable 5 Using GVS5H Harnessgithub.com

    ▲ 5pggues20d💬 1savesaved
14. 14
    

    newTuning a Local Coding Agent: Oh My Pi and Qwen3.8-27B on Two RTX 3090sdoug.sh

    ▲ 5dougcalobrisi21d💬 1savesaved
15. 15
    

    newPeer-to-peer LLM inference in browser tabs, Qwen 3.8 27Bgithub.com

    ▲ 5Nehanth29d💬 1savesaved
16. 16
    

    newStrata: Qwen3.8-Flash-Next (125B Moe) on a 8GB+ Nvidia GPUgithub.com

    ▲ 5simonpure8ddiscusssavesaved
17. 17
    

    newAsk HN: Which harness do you use with Qwen3.8 27B? And why?

    ▲ 4grigio13d💬 10savesaved
18. 18
    

    newUnattended 24hr Full-Stack Build: Qwen 3.8 27B on a Single RTX 5090github.com

    ▲ 4anglepoiselife15d💬 1savesaved
19. 19
    

    newMining Qwen 3.8 reasoning trace for prompt/skill evaluationolegivye.com

    ▲ 4olekrk25ddiscusssavesaved
20. 20
    

    newQwen3.8-Flash-Next Is on TensorFold with Speed Boostsx.com

    ▲ 3soltanov7d💬 2savesaved
21. 21
    

    newThinkingCap-Qwen3.8-27B: Less Thinking, Similar Accuracykaitchup.substack.com

    ▲ 3theanonymousone10d💬 1savesaved
22. 22
    

    newHalogen – Fastest Way to Run Qwen3.8-Flash-Next on AMD Strix Halogithub.com

    ▲ 3rdslw6ddiscusssavesaved
23. 23
    

    newSplash Engine – The Fastest Local Qwen3.8 on Apple Siliconlmstudio.ai

    ▲ 3sebiw17ddiscusssavesaved
24. 24
    

    newServe Qwen3 2.4T-Parameter Model with Configurable Reasoning on Nvidia GB300developer.nvidia.com

    ▲ 3backrun26ddiscusssavesaved
25. 25
    

    newJev-serve: Run latest qwen3.8, all LLM models, mlx-supportedgithub.com

    ▲ 2rreinold213d💬 2savesaved
26. 26
    

    newPewDiePie uncensors Qwen3.5-9B with Hereticwww.tomshardware.com

    ▲ 2alanwreath4d💬 1savesaved
27. 27
    

    newJeff-Code: a 0.8B model makes Qwen 3.8-27B coding 47% faster, same pass rategithub.com

    ▲ 2firelex36hdiscusssavesaved
28. 28
    

    newFull-fabric VHDL LLM inference engine. Runs Qwen3.5-class transformer inferencegithub.com

    ▲ 2jacquesm2ddiscusssavesaved
29. 29
    

    newGufo-Qwen3.6-35B-A3B-Q6dense – 3095tok/s prefill; 190 tok/s decode on Strix Halogithub.com

    ▲ 2nubela4ddiscusssavesaved
30. 30
    

    newShow HN: Moe Routing Atlas – which experts fire in Qwen3.5/3.6 35B-A3Bgslaller.github.io

    ▲ 2gslaller7ddiscusssavesaved

· edit