Gemini 1.5 Flash worth drop, tuning rollout full and enhancements to Gemini API and Google AI Studio
Final week, we launched an experimental up to date model of Gemini 1.5 Pro (0801) that ranked #1 on the LMSYS leaderboard for each textual content and multi-modal queries. We had been so excited by the speedy response to this mannequin that we raised the boundaries to check with it. We may have extra updates quickly.
Immediately, we’re asserting a sequence of enhancements throughout AI Studio and the Gemini API:
- Important discount in prices for Gemini 1.5 Flash, with enter token prices lowering by 78% and output token prices lowering by 71%
- Increasing the Gemini API to assist queries in 100+ further languages
- Expanded AI Studio entry for Google workspace clients
- Revamped documentation UI and API reference and extra!
Gemini 1.5 Flash worth lower
1.5 Flash is our hottest Gemini mannequin amongst builders who need to construct excessive quantity, low latency use circumstances similar to summarization, categorization, multi-modal understanding and extra. To make this mannequin much more inexpensive, as of August 12, we’re decreasing the enter worth by 78% to $0.075/1 million tokens and the output worth by 71% to $0.3/1 million tokens for prompts below 128K tokens (cascading the reductions throughout the >128K tokens tier in addition to caching). With these costs and instruments like context caching, builders ought to see main price financial savings when constructing with Gemini 1.5 Flash’s long context and multimodal capabilities.
Expanded Gemini API language availability
We’re increasing language understanding for each Gemini 1.5 Professional and Flash fashions to cowl greater than 100 languages so builders throughout the globe can now immediate and obtain outputs within the language of their selection. This could get rid of mannequin “language” block finish reasons by way of the Gemini API.
Google AI Studio entry for Google Workspace
Google Workspace customers can now entry Google AI Studio with out having to allow any further settings by default, unlocking frictionless entry for tens of millions of customers. Account admins will still have control to handle AI Studio entry.
Gemini 1.5 Flash tuning rollout now full
We’ve now rolled out Gemini 1.5 Flash text tuning to all builders by way of the Gemini API and Google AI Studio. Tuning allows builders to customise base fashions and enhance efficiency for duties by offering the mannequin further information. This helps cut back the context dimension of prompts, reduces latency and in some circumstances price, whereas additionally growing the accuracy of the mannequin on duties.
Improved developer documentation
Our developer documentation is core to the expertise of constructing with the Gemini API. We not too long ago launched a sequence of enhancements, up to date the content material, navigation, feel and appear, and launched a revamped API reference.
We’ve many extra enhancements to the documentation coming quickly so please proceed to ship us suggestions!
PDF Imaginative and prescient and Textual content understanding
The Gemini API and AI Studio now assist PDF understanding by means of each textual content and imaginative and prescient. In case your PDF consists of graphs, photos, or different non-text visible content material, the mannequin makes use of native multi-modal capabilities to course of the PDF. You may do this out by way of Google AI Studio or within the Gemini API.
Google AI Studio enhancements
Over the previous few weeks, we’ve got launched many enhancements to AI Studio, together with overhauling keyboard shortcuts, permitting dragging and dropping photos into the UI, decreased loading time by ~50%, added immediate strategies, and way more!
Builders are on the coronary heart of all our work on the Gemini API and Google AI Studio, so maintain constructing and sharing your suggestions with us by way of the Gemini API Developer Forum.
