2026-04-11 22:26 UTC
I have tried the new Gemma 4 Apache Licensed models e4b (9.6GB) and 26b (18GB)...
I've set up #Continue with Ollama on my VSCode.
They are very bad. Both models. I had hoped for maybe a slower response time, but better quality. They respond quite fast on my M1 MacBook with 32Gb of RAM, but the code quality is very poor.
#AI #Gemma #Ollama
Replies (1)
-
@dev@ricard.social 2026-04-11 22:26
On several occasions, what the agent returns in the chat is not what it has done in the files. Quite unusable. I knew I was spoiled, being accustomed to the Claude Sonnet models while at work, but I expected more from the Gemma 4 models. Maybe they are just not for coding, but general-purpose? #AI #Gemma #Ollama