Gemini 3 Flash hits 1M context at $0.50 – new default fast brain

Google’s Gemini 3 Flash is the first “fast” model in a while that actually feels frontier-class. You get a 1M-token context window, 64K outputs, and full multimodality at $0.50 per 1M input tokens and $3 per 1M output, with dynamic “thinking” levels baked into that price. Google is confident enough to swap it in as the default Fast and Thinking brain in the Gemini app and Search, while context caching chops repeated-prefix cost to 10% and batch jobs can see up to 50% savings.

The ecosystem read the memo instantly. Cursor, Cline, Warp, Zed, Antigravity, Perplexity, OpenRouter, Ollama and a slew of agent stacks all wired 3 Flash in as the new quick-but-smart default for coding, repo planning, and search-heavy workflows. Benchmarks back the shift: 78% on SWE-bench Verified, state-of-the-art on MMMU-Pro, a 71 score on the Artificial Analysis Index, and around 6× more successful BrowserUse web tasks per dollar than Claude Sonnet. Box AI saw document-extraction recall jump from 74% to 84% by swapping out 2.5 Flash.

There’s a catch: hallucination rates north of 90%, “lazy” short answers without prompt pressure, and day-one jailbreaks leaking MDMA recipes and runaway chain-of-thought. Treat Gemini 3 Flash as your new default workhorse—but only behind strong verification, safety filters, and house prompts that force it to think before it talks.