At its I/O developer convention, Google on Tuesday unveiled two new AI fashions: Gemini 3.5 Flash, the most recent mannequin in its Gemini sequence, and Gemini Omni Flash, a brand new multimodal mannequin that may “create something from any enter,” as Google places it.
Gemini 3.5 Flash
Gemini 3.5 Flash is the primary mannequin within the Gemini 3.5 sequence. A Professional model remains to be within the works and may launch subsequent month, however even 3.5 Flash bests the present 3.1 Professional mannequin in most benchmarks.
In TerminalBench 2.1, for instance, 3.1 Professional presently scores 70.3% when utilizing the Gemini CLI to resolve coding issues, whereas 3.5 Flash scores 76.2%.
That’s not fairly up there with OpenAI’s GPT 5.5, however for a Flash mannequin, that’s a really strong efficiency.
The brand new Flash mannequin will get comparable outcomes that outperform 3.1 Professional in different benchmarks, together with GDPval-AA (1656 Elo vs 1314), MCP Atlas (83.6% vs 78.2%), and CharXiv reasoning (84.2%).
However what’s possibly much more fascinating is that it’s aggressive — and typically even beats — flagship fashions like OpenAI’s GPT-5.5 and Anthropic’s Opus 4.7 in fairly a couple of benchmarks, too.
That’s very true for instrument utilization benchmarks, and as Google CEO Sundar Pichai famous in a press briefing forward of as we speak’s launch, Gemini 3.5 Flash is “a primary in a sequence of fashions combining frontier intelligence with actions.”
He famous that Flash is near the most effective frontier fashions, but additionally very quick. Synthetic Evaluation places it simply behind the frontier fashions of OpenAI and Anthropic, however at considerably sooner token per second speeds (near 280 tokens per second vs. round 60 or 70 for GPT-5.5 and Opus 4.7).
“What’s wonderful about Flash is the way it delivers frontier-level capabilities at lower than half the value, in some circumstances nearly a 3rd of the value of comparable frontier fashions,” Pichai famous.
Google notes that 3.5 Flash is very robust in terms of operating long-horizon agentic duties, together with agentic coding. That’s additionally why this mannequin is on the core of Gemini Spark, the brand new private AI agent Google is launching at I/O (and which is presently solely accessible to trusted testers).
Given the capabilities of the Flash mannequin, the Professional mannequin will probably be a minimum of as succesful as comparable fashions from OpenAI and Anthropic — and sure greatest them in a minimum of some benchmarks.
Gemini 3.5 Flash availability
Gemini 3.5 Flash is now accessible by way of the Gemini API in Google AI Studio and Android Studio, Gemini Enterprise Agent Platform (aka Vertex AI), and Gemini Enterprise, in addition to in Google Antigravity.
For shoppers, it is usually accessible within the Gemini app and AI Mode in Google Search.
Gemini Omni
Gemini Omni is a little bit of a unique mannequin. The Gemini fashions have been, to a point, all the time meant to be multimodal, however Omni takes this fairly a bit additional. In its present iteration, it’s a bit extra like Veo, Google’s generative mannequin for creating movies, however over time, it can additionally assist photos and audio.
So regardless that Google says Omni can “create something from any enter,” it’s beginning with solely video for now. That has confirmed to be an space the place there was lots of progress within the final 12 months or so, and Omni brings most of the capabilities of what customers now anticipate from picture fashions to video.
Omni, which, like Gemini 3.5 is barely launching with a Flash mannequin for now, lets customers change particular issues in a video, for instance, and fully reimagine a shot by including new characters and objects, or altering the surroundings, angle, and magnificence. Google says it could possibly do that “with out ever dropping the thread of your unique scene.”
As Google stresses (and different frontier labs are inclined to argue the identical about their video fashions), Omni’s world mannequin has an ‘intuitive’ understanding of gravity, kinetic power, and fluid dynamics, which ought to make for life like scenes.
As a result of it’s multimodal (or will probably be quickly), Omni can take photos, textual content, video, and audio (or any mixture of these) as its enter to construct the ultimate scene.
A digital avatar on your security (and people round you)
Generative video lends itself to deepfakes and misinformation campaigns. Google says it’s “dedicated to growing AI responsibly and we have now clear insurance policies to guard customers from hurt and governing the usage of our AI instruments.” In follow, this implies you’ll be able to presently create movies with your individual voice and an avatar of your likeness.
“Past the avatar characteristic, when it comes to enhancing movies to alter audio and speech, we’re nonetheless working to check this and higher perceive how we are able to convey this functionality to customers responsibly,” Google says.
All movies created with Gemini Omni will embody Google’s SynthID watermark.
YOUTUBE.COM/THENEWSTACK
Tech moves fast, don’t miss an episode. Subscribe to our YouTube
channel to stream all our podcasts, interviews, demos, and more.
