MOUNTAIN VIEW, Calif. — Google on Tuesday used the keynote of its annual I/O developer conference to ship a coordinated three-product Gemini update — the lightweight Gemini 3.5 Flash, the multimodal Gemini Omni, and a general-purpose agent the company has named Spark.
The release is, in scope, the most expansive single Gemini moment the laboratory has staged. It places Google in direct rhetorical competition with each of the three product directions its rivals at Anthropic, OpenAI, and Microsoft have foregrounded in the past month: enterprise model adoption, default-model reliability, and agent-first infrastructure, respectively.
Gemini 3.5 Flash, the laboratory said, offers cutting-edge capabilities at roughly half — and in some cases close to one-third — the price of comparable frontier models. The variant is the consumer-facing successor to the 3.1 Flash-Lite generally-available release that landed earlier in the month.
Gemini Omni is the architecturally more ambitious of the three. Google described it as a leap forward in world understanding, multimodality and editing, a model designed to generate any output from any input, starting with video. The framing — and the order of the bullet points in the keynote — appears to be an explicit response to OpenAI’s release of GPT-5.5 Instant on May 5, which focused on reliability rather than expanding the modality envelope.
Gemini Spark, the third product, is the agent. Google described it as a new general-purpose artificial-intelligence agent capable of reasoning across information in connected applications. The product launched in beta, available first to trusted testers and subscribers to Google AI Ultra, the company’s most expensive consumer tier.
Earlier in the month — on May 12 — Google had separately announced that Gemini Intelligence would be coming to Android devices, beginning with select Samsung and Google phones over the summer and expanding to a broader device set later this year. The Android disclosure, by accounts of executives present, was deliberately staged in advance of the I/O announcements to position Gemini Intelligence as the user-facing surface that Spark and Omni would feed into.
The three-product approach is consistent with Google’s recent product cadence: a model, a modality push, and an agent surface, shipped together rather than sequenced. The competitive implication, for the broader market, is that Google intends to compete on all three frontiers simultaneously rather than picking one — a posture that the company can afford in part because, unlike its competitors, it owns the operating system, the device fleet, and the data center the products will ultimately run on.