2026-04-23 13:58 UTC
Is the #LLM race actually a race to the bottom? In the short time I've been tracking model development, the jump in what's possible on consumer hardware has been impressive.
Every other week, we see a new model that does more with lighter weights and fewer parameters.
#AI #qwen #kimi #gemma4 #selfhosting
Replies (2)
-
@perpetuum_mobile@mastodon.social 2026-04-23 14:08
@mamba@mstdn.ca Gemma4:26B is sweet spot on my RTX3090: - 108 token/s - that thing is FAST - multimodal - you don't need second VLM model - you can variably turn reasoning on/off using reasoning_effort header in API calls - performance comparable and sometimes exceeding cloud Gemini-2.5-Flash New Qwen 3.6 27B is also very strong, the only downside : 3 times slower than Gemma4 Fully local AI on consumer HW finally caught up with cloud.
-
@luca@mastodon.luca-alloatti.eu 2026-04-27 08:16
@mamba@mstdn.ca I totally agree. Qwen3.6-27B is 15x smaller than Qwen3.5-397B-A17B and better at coding. I am running the new 27B model on my GPU and I am absolutely impressed.