It's been about 2 weeks completely using open weight's models for my development work flows and it's been a better, and more consistent experience than frontier models since the big retardation.
My current preference is GLM 5.1 and 5.0, but kimi-k 2.6 is fine. Qwen has left much to be desired imo. My favorite chat model currently is deepseek 4.
None of these are affordable to run locally, don't bother imo. I spend a lot of time using smaller models like ~30b range for "privacy sensitive" work and it's worth trying if you have ~32GB of vram available.
All inference providers currently suck in different ways, which I'm sure y'all have experience with so we've been testing the bare minimum hardware setups able to sustain usable quality output tokens, but that's still a very fine line.
ChipTuner
npub1qd…3fqm7
2026-05-17 18:36:45 UTC
Author Public Key
npub1qdjn8j4gwgmkj3k5un775nq6q3q7mguv5tvajstmkdsqdja2havq03fqm7Published at
2026-05-17 18:36:45 UTCEvent JSON
{
"id": "8366adffac751dab50cd0498211b65e59a290babbaffaa0d09beddb2c338883c",
"pubkey": "036533caa872376946d4e4fdea4c1a0441eda38ca2d9d9417bb36006cbaabf58",
"created_at": 1779043005,
"kind": 1,
"tags": [],
"content": "It's been about 2 weeks completely using open weight's models for my development work flows and it's been a better, and more consistent experience than frontier models since the big retardation.\n\nMy current preference is GLM 5.1 and 5.0, but kimi-k 2.6 is fine. Qwen has left much to be desired imo. My favorite chat model currently is deepseek 4. \n\nNone of these are affordable to run locally, don't bother imo. I spend a lot of time using smaller models like ~30b range for \"privacy sensitive\" work and it's worth trying if you have ~32GB of vram available. \n\nAll inference providers currently suck in different ways, which I'm sure y'all have experience with so we've been testing the bare minimum hardware setups able to sustain usable quality output tokens, but that's still a very fine line.",
"sig": "d2546f40326823c2fa63ea668090099068eec500013530727ca6c362e2217b5dd2c48d2decdbd34558fce79e486c9089a4444dec6b34c09ca6833ad80ee7cdd8"
}