Google's Gemma4 gets multiple bug fixes in 24 hours

Google just pushed out fixes for Gemma4, their latest AI model, and they're actually pretty important. First up: a reasoning budget fix that got merged into llama.cpp (that's the popular open-source tool people use to run these models locally). But here's the bigger deal—Google released new chat templates specifically to fix how the model handles tool calling, which basically means it can now properly use external tools and functions. They've updated both the 31B and 27B versions. If you're running Gemma4 locally or building apps with it, these fixes should make the model way more reliable, especially if you're trying to get it to actually do things beyond just chatting.