Hackernews posts about Qwen 3

1. 01
   

   newQwen 3.8 27B on my laptop can interpret model weights, toosimonwillison.net

   ▲ 736theanonymousone3d💬 214savesaved
2. 02
   

   newQwenputer: A functioning, deterministic RISC-V CPU inside Qwen3.5divagr.com

   ▲ 708csmantle9d💬 147savesaved
3. 03
   

   newMemdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hardgithub.com

   ▲ 465guregodevo6d💬 232savesaved
4. 04
   

   newQwen3-Coder writes less secure code when told it's for a US personawww.hirundo.io

   ▲ 243mattifrank16d💬 118savesaved
5. 05
   

   newFive Qwen3.8-27B models match Fable 5 Using GVS5H Harnessgithub.com

   ▲ 196rzk13d💬 75savesaved
6. 06
   

   newShow HN: Qwen3-4B with the AI writing tics trained out100toolkit.com

   ▲ 158erikkannike18d💬 113savesaved
7. 07
   

   newQwen3.8-27B on the iPhone 18 Neural Engine at 15–70 tok/s (1.5 bpw)twitter.com

   ▲ 30anemll20d💬 12savesaved
8. 08
   

   newRunning Qwen3.8-Flash-Next NVFP4 at 184 tok/s on a single RTX 6000 Promeshapplied.com

   ▲ 29matteohorvath11d💬 6savesaved
9. 09
   

   newNeuron Explorer for Qwen3.5-4B: See every neuron at work, make targeted editsgithub.com

   ▲ 29Schlagbohrer11ddiscusssavesaved
10. 10
    

    newRunning Qwen3.8 at 255 tok/s on a single 5090meshapplied.com

    ▲ 15Ir0n24d💬 13savesaved
11. 11
    

    newLithos-metal: Qwen3.8-27B on 40-core M5 Max at 168 tokens/sgithub.com

    ▲ 11emp_21d💬 1savesaved
12. 12
    

    newKapitsa: New Apple Silicon Optimized Engine Runs Qwen 35B on 160 Tok/S on M4 Prowww.reddit.com

    ▲ 8fearandgreed17ddiscusssavesaved
13. 13
    

    newQwenputer: A functioning, deterministic RISC-V CPU inside Qwen3.5divagr.com

    ▲ 7schmorptron9ddiscusssavesaved
14. 14
    

    newAlibaba's Qwen3.8-Flash-Next model tried to upload a file to Beijing in a testwww.notebookcheck.net

    ▲ 5nh43215rgb11ddiscusssavesaved
15. 15
    

    newQwen3.8-27B at ~200 tok/s peak on an Apple M5 Maxtwitter.com

    ▲ 5rthng29ddiscusssavesaved
16. 16
    

    newArete-27B: Qwen3.8-27B, post-trained to reinforce Western valuesnousresearch.com

    ▲ 4pretext2h💬 12savesaved
17. 17
    

    newQwenGPU: A functioning, deterministic SIMT GPU inside Qwen3.5twitter.com

    ▲ 4vixhal4ddiscusssavesaved
18. 18
    

    new2x Tesla p100s, q6_k quant, Qwen 3.8 27B ~60tpsgithub.com

    ▲ 3Kmic6819d💬 2savesaved
19. 19
    

    newCoinbase Built, Evaluated, and Post-Trained a Fraud Agent on Qwen3.5-9Bwww.coinbase.com

    ▲ 37777777phil24d💬 1savesaved
20. 20
    

    newA steerable motivation concept influencing persistence in Qwen-3.5-27Bthebitsarealive.substack.com

    ▲ 3gfysfm11ddiscusssavesaved
21. 21
    

    newQwen3.8-Omni: Towards Native Omni-Modal Agentsarxiv.org

    ▲ 3teleforce24ddiscusssavesaved
22. 22
    

    newUnderdog Saluki: Qwen3.8-27B in under 8GB with tool calling intacthuggingface.co

    ▲ 3acossta28ddiscusssavesaved
23. 23
    

    newBias in Qwen3-Coder Modelwww.schneier.com

    ▲ 2bretpiatt10d💬 1savesaved
24. 24
    

    newCustom Voices with Qwen3-TTSlucumr.pocoo.org

    ▲ 2mattvr3ddiscusssavesaved
25. 25
    

    newDFlash2 drafter for Qwen3.6 trained on coding workloadshuggingface.co

    ▲ 2ryanbrosas_cb4ddiscusssavesaved
26. 26
    

    newQwen 3.8 invents URLs with real key ID, that's why I think it's no backdoorwww.xda-developers.com

    ▲ 2manveru9ddiscusssavesaved
27. 27
    

    newVibeSys Builds a Qwen3.5-397B Engine on MI300A, 2.3× Tuned SGLangsyfi.cs.washington.edu

    ▲ 2matt_d28ddiscusssavesaved
28. 28
    

    newNanbeige/Nanbeige4.3-33B - stronger coder than Qwen 3.8huggingface.co

    ▲ 1rene_d4d💬 3savesaved
29. 29
    

    newTested 18 new Local LLMs on two RTX 5060 Ti cards: none replaced Qwen3.8-27Bsiimtolk.github.io

    ▲ 1TheQuantumPhant15d💬 1savesaved
30. 30
    

    newVibeSys on Trainium: Agents Build/Tune Qwen3.8 Serving Engines on AWS Trainium2syfi.cs.washington.edu

    ▲ 1matt_d35hdiscusssavesaved

· edit