Not really. Qwen 3.8 is a dense model. At Q4 mentioned here ...

npub10nev22800j2q5ncjcpynyqakyrhsg49y046ucyzlx3jj5mns7mmsv6spqj

hex

33435057da751aec85267433e661257504886a6384a8a8748aef6e4181be0c9f

nevent

nevent1qqsrxs6s2ld82xhvs5n8gvlxvyjh2pygdf3cf29gwj9w7mjpsxlqe8cprpmhxue69uhhyetvv9ujuem4d36kwatvw5hx6mm9qgs8euk99rhhe9q2fufvqjfjqwmzpmcy2jj86awvzp0ngef2dec0dac9ac8m3

Kind-1 (TextNote)

2026-08-20T11:49:24Z

↳ Reply to TheGrinder (npub1de6l09erjl9r990q7n9ql0rwh8x8n059ht7a267n0q3qe28wua8q20q0sd)

"NVIDIA RTX GPUs Bring Frontier Quality Qwen3.8-27B Coding Agents to Local AI" https://blogs.nvidia.com/blog/local-ai-open-source-models-agents-nemotr...

Not really. Qwen 3.8 is a dense model. At Q4 mentioned here it's pretty shitty. And still needs a lot of vram even at q4.

If it was MoE, this would be closer to a small revolution

Raw JSON

{
  "kind": 1,
  "id": "33435057da751aec85267433e661257504886a6384a8a8748aef6e4181be0c9f",
  "pubkey": "7cf2c528ef7c940a4f12c0493203b620ef0454a47d75cc105f34652a6e70f6f7",
  "created_at": 1787226564,
  "tags": [
    [
      "e",
      "9071ecc3d1f3e013a48bc8843a4d67d73b3abbe28ebc184e1d51e554140a268e",
      "wss://relay.primal.net/",
      "root",
      "6e75f7972397ca3295e0f4ca0fbc6eb9cc79be85bafdd56bd378220ca8eee74e"
    ],
    [
      "p",
      "6e75f7972397ca3295e0f4ca0fbc6eb9cc79be85bafdd56bd378220ca8eee74e",
      "wss://relay.nostrplebs.com/"
    ],
    [
      "client",
      "Amethyst"
    ]
  ],
  "content": "Not really. Qwen 3.8 is a dense model. At Q4 mentioned here it's pretty shitty. And still needs a lot of vram even at q4.\n\nIf it was MoE, this would be closer to a small revolution",
  "sig": "7d85e67708e2a6fb46509c5f985e36ce92f01636668dc79cfcd4d6cf56dfb561f86589fbbc97e9fdfc6e07048b837b7e911ac2060efd2bd8954a5568277f759c"
}