Comment on: Google Gemma 4

The agentic workflow angle is the interesting part for me. Most open models get benchmarked on reasoning and coding, but the harder question for production use is how they handle multi-step tasks where the model needs to recover from partial failures.Running Claude Code agents in parallel - local inference becomes appealing but reliability in long workflows is still the blocker. Anyone tested Gemma 4 on tasks with 10+ tool calls?