Google I/O Unveils Gemini Omni, Gemini 3.5 Flash, and Spark: A Full‑Scale AI Leap

Google I/O 2026 showcased a series of AI breakthroughs. The headline announcement was Gemini Omni, described as a truly "all‑purpose" model that accepts arbitrary combinations of images, audio, video, and text and generates high‑quality video output. In a live demo, a prompt "Explain protein folding with clay animation" produced a scientifically accurate animation of amino‑acid chains forming α‑helices and β‑sheets, and another demo mapped each English letter to a distinct object (e.g., C → Capybara, D → disco ball, L → lava lamp), demonstrating genuine multimodal understanding rather than simple collage.

Gemini 3.5 Flash was presented as the next‑generation flagship model. According to the presentation, it surpasses Gemini 3.1 Pro on all benchmark suites: Terminal‑Bench 2.1 (coding) 76.2 %, GDPval‑AA (real‑world agent tasks) 1656 Elo, MCP Atlas (large‑scale tool use) 83.6 %, and CharXiv Reasoning (multimodal comprehension) 84.2 %. The model also achieved 289 tokens / s, roughly four times faster than competing frontier models, and its speed places it in a unique performance quadrant.

The event also introduced Antigravity 2.0, an upgraded agent development platform that transforms from an IDE into a standalone desktop application. Antigravity enables dynamic sub‑agent creation, asynchronous task scheduling, and new slash commands such as /goal, /grill‑me, and /browser. In a live demonstration, 93 parallel sub‑agents issued over 15 000 model calls, processed 2.6 billion tokens, and within 12 hours turned an empty project into a fully functional OS kernel, writing the scheduler, memory manager, and file system without human‑written code. The API cost for this workload was reported to be under $1 000.

Gemini Spark, the third major release, is positioned as a 7 × 24 h personal AI agent. Powered by Gemini 3.5 + Antigravity, Spark can draft emails by pulling context from Gmail, Docs, and chat history, apply a custom "ghostwriter" skill to match the user's tone, and orchestrate complex workflows such as planning a neighborhood block party—automatically creating a Google Sheet RSVP tracker, generating a Slides deck, and sending follow‑up messages—all without the user opening any app. Voice input was demonstrated by issuing three tasks in a single utterance, which Spark split into parallel threads and executed concurrently.

Pricing details were also disclosed: the AI Ultra subscription now costs $100 / month for Spark beta access, with the top‑tier plan reduced from $250 to $200. Gemini 3.5 Flash will be the default model for Gemini App, Google Search AI mode, and the Gemini API, while Gemini 3.5 Pro is slated for release next month.

Overall, the I/O announcements illustrate a convergence of full‑modal understanding, generative capabilities, and always‑on agents, suggesting that the technical barriers to artificial superintelligence have largely been removed and the remaining challenge is deployment speed.