Gemini 3.6 Flash ships: 17% cheaper output, DeepSWE up to 49%
Google's newest Flash is GA across AI Studio, Vertex and Antigravity, cheaper than 3.5 Flash, and ships beside a low-cost Flash-Lite and a Gemini 4 teaser.
Copy markdown
GA now, tuned for agents
Gemini 3.6 Flash is generally available through the Gemini API, Google AI Studio, Vertex AI, Antigravity 2.0 and Android Studio, with a 1M-token context and a March 2026 knowledge cutoff. Google says it takes fewer reasoning steps and tool calls to finish multi-step workflows.
Output is ~17% cheaper
Output drops to $7.50 per 1M tokens (from 3.5 Flash's $9.00), input holds at $1.50, and cached input is $0.15. Because it also emits roughly 17% fewer output tokens per task, your real bill falls further than the sticker price suggests.
Real coding + computer-use gains
DeepSWE climbs to 49% (from 37%), MLE-Bench to 63.9% (from 49.7%), OSWorld-Verified to 83.0% (from 78.4%), and GDPval-AA to 1421 Elo (from 1349). Google claims fewer unwanted code edits and shorter execution loops.
Flash-Lite for high-volume jobs
A cheaper Gemini 3.5 Flash-Lite ships alongside at $0.30/$2.50 per 1M, aimed at document processing and agentic search. Terminal-Bench 2.1 jumps to 54% (from 31%) and long-context to 72.2% (from 60.1%).
Flash Cyber stays gated
A security-focused Gemini 3.5 Flash Cyber for vulnerability detection is limited-pilot only, open to governments and trusted partners — not callable by individual builders yet.
Gemini 4 is now pre-training
DeepMind confirmed it has begun 'the most ambitious pre-training run yet' for Gemini 4. Nothing shippable, but a signal the fast Flash cadence is a bridge to the next frontier model.