What's going on with Gemini?
a day ago
- Google's Gemini models trail behind Anthropic and OpenAI in frontier intelligence, with Gemini 3.1 Pro falling short of leading models like GPT-5.5 and Opus 4.8.
- Gemini 3.5 Flash offers impressive speed (206 t/s) but has a 3x price increase and underwhelming coding benchmarks, making its market position unclear.
- Google's custom TPU hardware and internal model-hardware collaboration provide structural advantages for inference efficiency.
- Google lacks a coherent coding agent strategy, with multiple overlapping tools (Antigravity, Jules, Gemini Code Assist, Gemini CLI, AI Studio) and weak developer adoption compared to Anthropic's Claude Code and OpenAI's Codex.
- Google's bespoke internal software workflows may hinder its ability to develop effective agentic tools for the broader industry.