Lightweight models are coming for the flagship crown. Google announced Gemini 3.5 Flash at I/O 2026, a lightweight flagship model rolling out globally across Gemini apps, AI search, and the Antigravity developer tool. The more forward-looking Gemini 3.5 Pro is due next month.
Third-party benchmarkers at Artificial Analysis clocked Gemini 3.5 Flash at 289 tokens per second — nearly four times faster than GPT-5.5's 71 tokens/s and Claude Opus 4.7's 67 tokens/s. It also doubles the speed of its midrange predecessor, Gemini 3.1 Pro (135 tokens/s).
But speed didn't come at the cost of control. On the MCP Atlas test for multi-step tool use, Gemini 3.5 Flash scored 83.6%, beating GPT-5.5 (75.3%) and Claude Opus 4.7 (79.1%). It also outperformed previous pro models in Terminal-Bench 2.1, with 76.2% versus Gemini 3.1 Pro's 70.3% and Claude Opus 4.7's 66.1%.
The model supports a 1 million token context window and scored 1656 on the GDPval-AA test for long-term task stability, up from 1314 on the previous generation.
The model is available now. Consumers can use it via the web and Search AI mode; developers can hit the API through Google AI Studio, Android Studio, and Antigravity; enterprises can deploy via the agentic platform. Standard API pricing is $1.50 per million input tokens and $9.00 per million output tokens.