Because the launch of the Gemini 2.0 Flash mannequin household, builders are discovering new use instances for this extremely environment friendly household of fashions. Gemini 2.0 Flash provides stronger efficiency over 1.5 Flash and 1.5 Professional, plus simplified pricing that makes our 1 million token context window extra reasonably priced.
At this time, Gemini 2.0 Flash-Lite is now typically out there within the Gemini API for manufacturing use in Google AI Studio and for enterprise clients on Vertex AI. 2.0 Flash-Lite provides improved efficiency over 1.5 Flash throughout reasoning, multimodal, math and factuality benchmarks. For tasks that require lengthy context home windows, 2.0 Flash-Lite is an much more cost-effective resolution, with simplified pricing for prompts greater than 128K tokens.
Builders are already leveraging the velocity, effectivity, and cost-effectiveness of the two.0 Flash household to construct unbelievable functions. Listed below are a couple of examples:
1. Voice AI
Constructing efficient conversational AI, significantly voice assistants, requires each velocity and accuracy. A quick Time-to-First-Token (TTFT) is crucial for making a pure, responsive really feel, alongside the power to deal with complicated directions and work together with different methods through perform calling.
Daily is leveraging Gemini 2.0 Flash-Lite to assist builders create cutting-edge voice AI experiences. Utilizing their open-source, vendor agnostic Pipecat framework for voice and multimodal conversational brokers, Every day has created a system instruction code demo to reliably detect voicemail methods and tailor messages accordingly.
Gemini 2.0 Flash-Lite, with the above system instruction, performs considerably higher than present specialised industrial fashions for detecting voicemail.
2. Information analytics
Dawn is revolutionizing how engineering groups monitor their AI merchandise in manufacturing by offering deep, significant insights powered by Gemini 2.0 Flash. Daybreak’s “semantic monitoring” pipeline permits engineering groups to immediately search large streams of consumer interactions to seek out any habits they’re searching for—like consumer frustration, dialog size, and consumer suggestions—and repeatedly observe them as ongoing points or matters to determine anomalies and hidden issues in manufacturing.
With Gemini 2.0 Flash’s simplified pricing, dependable structured outputs, and prolonged context capabilities, Daybreak was in a position considerably cut back search occasions (from hours to only beneath a minute) by switching fashions, minimize prices by greater than 90%, and see elevated reliability throughout evals and manufacturing monitoring.
Gemini 2.0 Flash makes Daybreak’s semantic monitoring quicker, extra dependable, and price efficient.
3. Video enhancing
Mosaic is reworking complicated, time-consuming video enhancing duties with a brand new, agentic paradigm that makes use of Gemini 2.0 Flash. Their resolution incorporates multimodal enhancing brokers that use Gemini 2.0 Flash’s long-context capabilities to speed up mundane video enhancing duties from hours to seconds so you are able to do issues like clip YouTube Shorts from any a part of an extended type video with only a immediate.
The brand new simplified pricing for Gemini 2.0 Flash of $0.10 per 1 million enter tokens in Google AI Studio makes enormous context home windows 33% extra reasonably priced, opening up new potentialities for AI-driven video enhancing workflows.
Utilizing Gemini 2.0 Flash, Mosaic’s agentic workflow cuts and edits a YouTube Brief from a current episode of Launch Notes.
Begin constructing with Gemini 2.0 Flash and a couple of.0 Flash-Lite
We’re excited by what the Gemini 2.0 Flash household of fashions is enabling for builders like Daily.co, Mosaic, and Dawn. Whether or not you are engaged on voice assistants, video enhancing instruments, or one thing utterly new, we hope the Gemini 2.0 Flash household supplies the efficiency and affordability you want. Begin constructing at the moment in Google AI Studio.
