What (I think) makes Gemini 3 Flash so good and fast

kept by eddie dec 23, 25

Gemini 3 Flash likely uses a massive 1.2 trillion-parameter ultra-sparse MoE architecture, activating only 5–30 billion parameters per inference to deliver high intelligence at low cost, though it suffers from token bloat and a 91% hallucination rate on refusals.

Surpassing Frontier Performance with Fusion — OpenRouter Blog
OpenRouter's Fusion, which synthesizes outputs from multiple language models via a judge model, outperformed individual frontier models including GPT-5.5 and Claude Opus 4.8 on deep research tasks, with a budget panel achieving near-frontier performance at half the cost.
also on Gemini 3 Flash, #ai-research
Gemini 3 Flash: Frontier intelligence built for speed
Google announced Gemini 3 Flash, a frontier-intelligence model optimized for speed and cost, outperforming prior Pro models on key benchmarks, now available to developers.
also on Gemini 3 Flash, #generative-ai, #ai
Use multiple models - by Nathan Lambert - Interconnects
To get the most out of AI in 2026, the author recommends using multiple models for different tasks as each model has jagged capabilities.
also on Gemini 3 Pro, #generative-ai, #ai
Google adding seamless transition from AI Overviews to AI Mode
Google is testing a seamless transition from AI Overviews to AI Mode on mobile, and expanding Gemini 3 Pro to 120 countries.
also on Gemini 3 Pro, #generative-ai, #ai
Frontier labs don’t use most AI compute (yet) - by Josh You
Epoch AI estimates frontier labs use less than half of global AI compute, but OpenAI and Anthropic may soon dominate, requiring economic transformation to sustain scaling.
also on Google DeepMind, #scaling-laws, #ai