Gemma 4 31B hits tool-call bugs
Google's newly released Gemma 4 31B flagship is reportedly failing basic tool calls in early LocalLLaMA tests. Despite high reasoning benchmarks, community reports point to inference-engine compatibility issues with llama-cpp and the new Apache 2.0 licensed model.
Gemma 4 31B's promising agentic performance is being bottlenecked by day-one implementation bugs in popular local inference tools. High benchmark scores fail to translate into reliability due to inference engine issues handling the new Gemma 4 tool-calling format in llama-cpp. As Google's first truly open-licensed flagship, these early local-hosting hurdles highlight a critical gap between raw model capability and ecosystem readiness.
DISCOVERED
9d ago
2026-04-03
PUBLISHED
9d ago
2026-04-02
RELEVANCE
AUTHOR
ForsookComparison