Google is out with a new model — Gemini 3.6 Flash — aimed squarely at coding, knowledge work, and multimodal agents. The company says the model needs fewer reasoning steps and tool calls to complete multi-step tasks.
Independent analysis from Artificial Analysis shows it uses 17% fewer output tokens than Gemini 3.5 Flash. On the API side, input pricing stays at $1.50 per million tokens, while output drops from $9 to $7.50 per million tokens. With both token usage and per-token cost falling, running long agent tasks just got cheaper.
Performance is also up. On DeepSWE, scores rose from 37% to 49%; on MLE Bench from 49.7% to 63.9%; and on OSWorld-Verified from 78.4% to 83%.
Google also introduced a cheaper model: Gemini 3.5 Flash-Lite, which cranks out 350 tokens per second. Its input and output prices are $0.30 and $2.50 per million tokens, respectively.
A cybersecurity-focused variant, Gemini 3.5 Flash Cyber, will enter limited testing, available only to government agencies and trusted partners.
And the next big thing is already in motion: Gemini 4 has started pre-training, which Google calls its most ambitious pre-training effort yet. Meanwhile, Gemini 3.5 Pro remains in partner testing and will roll out widely once ready.