Builders and prospects constructing manufacturing AI brokers require increased token effectivity, decrease latency, and extra dependable efficiency. Our Flash sequence of fashions are constructed to fulfill the candy spot of effectivity and high quality, permitting you to scale your agent workflows. We’re introducing a brand new Gemini mannequin primarily based on Gemini 3.5 Flash.
- 3.6 Flash: Our flagship mannequin for higher coding, data work, and multimodal efficiency. Based on artificial analysis index17% discount in output token utilization in comparison with 3.5 Flash, and a few benchmarks resembling DeepSWE. data curveas much as 65% is noticed, all with a decrease value per output token.
- 3.5 Flashlight: Our quickest and most cost-effective 3.5 class mannequin delivers 350 output tokens per second in accordance with the Synthetic Evaluation Index, considerably outperforming earlier technology Flash-Lite in agent workflows.
- 3.5 CodeMender’s Flash Cyber: Profitable cybersecurity functions require cautious orchestration of fashions in parallel with agent infrastructure. We’re introducing a brand new mannequin of extremely environment friendly and specialised cyber specialization mixed with the CodeMender code safety agent that gives cutting-edge and aggressive efficiency.
Along with right now’s launch, Gemini 3.5 Professional is presently being examined with companions and will likely be extensively obtainable when prepared. In parallel, our group is already centered on constructing next-generation fashions. We’ve begun our most bold pre-training for Gemini 4 so far and are enthusiastic about our progress.
3.6 Flash: Extra environment friendly and better high quality than 3.5 Flash.
Gemini 3.6 Flash is constructed instantly on developer and buyer suggestions from 3.5 Flash. 3.6 Flash not solely supplies a step-up in coding and data work, it does this whereas considerably growing token effectivity. For instance, the Synthetic Evaluation Index reveals that 3.6 flashes eat 17% fewer output tokens than 3.5 flashes. It additionally requires fewer inference steps and power calls to perform multi-step workflows.
This enhanced effectivity additionally comes at a lower cost than 3.5 flash. 3.6 Flash reduces the general value per agent activity at $150 per million enter tokens and $750 per million output tokens, making it cheaper to construct and run brokers.
