A Reddit user shared benchmark results showing Google's Gemma 4 31B (FP8) performing on par with Claude Sonnet 4.6 Medium. The custom evaluation harness tested complex tasks including Neo4j Cypher queries, entity extraction, agentic tool calling, Python coding, and multi-vector retrieval synthesis. This highlights how quantized mid-sized open-source models are closing the gap with leading proprietary frontier models.
Vercel announced in its latest Changelog that the Vercel Agent — its built-in AI-assisted agent tool — has gained a new automated installation capability…