Google I/O 2026 recap — Gemini 3.5 Flash ships; Omni, Spark, and a $100 AI Ultra tier

What actually shipped today

Five layers of news from the May 19 keynote, ordered by immediate user impact:

1. Gemini 3.5 Flash — generally available today

The new flagship Flash-tier model rolled out across the Gemini app, Search (AI Mode), Antigravity 2.0, and the Gemini API. Per Google’s benchmark disclosures:

• Surpasses Gemini 3.1 Pro on coding, agentic, and multimodal benchmarks
• 4x faster than other frontier models on output tokens per second (Google’s claim)
• Flash-tier cost and speed (substantially cheaper than Pro)

The implication: Gemini’s Flash tier in 2026 is producing what was Pro-tier capability a year ago. For developers running high-volume workloads on Gemini API, the cost-quality math improved materially today.

2. Gemini 3.5 Pro — preview window, launching next month

Pichai confirmed Gemini 3.5 Pro is in internal testing but won’t be broadly available until next month. The preview-then-ship cadence is the standard Google pattern, but worth tracking — Pro typically sits at the top of benchmark leaderboards when it lands.

For Gemini Pro subscribers, the implicit timeline: continue on Gemini 3 Pro through May; upgrade to 3.5 Pro automatically in June when it ships.

3. Gemini Omni — the video model confirmed

The Omni leak from May 11 was real. As announced today, Gemini Omni is a series of models (not a single model) that combines reasoning with creation. Confirmed capabilities:

• Multimodal input: image, audio, video, and text
• Output: editable video grounded in real-world knowledge
• Distribution: Gemini app, Google Flow, and YouTube Shorts
• Tier: AI Plus, AI Pro, and AI Ultra (paid tiers — not free)

Notably, Omni did not launch on the Gemini free tier today, which contradicts the pre-event prediction that Google would deploy the Nano-Banana-Pro distribution playbook (free for everyone). Instead, Omni sits behind a paywall — at least for now. The competitive read on Veo 3, Runway, and Kling needs to be revisited with this paid-only positioning in mind.

4. Gemini Spark — the personal agent

Spark is Google’s answer to the AI-agent push from Anthropic (May 5 finance-agent templates) and OpenAI (May 11 Deployment Company $4B services arm). Per the announcement:

• Personal agent that takes actions on your behalf
• Workspace-integrated — pulls context from Gmail, Docs, Calendar
• MCP (Model Context Protocol) support coming summer 2026 — allows third-party tools to be exposed to Spark as callable tools
• Availability: Google AI Ultra subscribers, US only, starting next week
• Plus a Daily Brief feature for AI Plus/Pro/Ultra rolling out today

The MCP support is the strategic detail. Anthropic created MCP and shipped it across Claude Code and the Anthropic ecosystem; Google adopting it for Spark means the agent-tool interoperability layer is becoming a cross-vendor standard rather than an Anthropic-only protocol.

5. Pricing restructure — the math actually changed

The most user-impacting announcement: Google AI Ultra is now $100/month (was $249.99). The previous $249.99 tier dropped to $200/month with same capabilities — meaning existing $250 subscribers get a $50/month price cut, and there’s now an additional $100 tier in the middle that didn’t exist before.

The new tier structure:

• Google AI Pro — $19.99/month (unchanged)
• Google AI Ultra (new $100 tier) — 5x Pro usage limits, Gemini Spark access
• Google AI Ultra ($200 tier) — formerly $249.99; same capabilities at lower price

Plus a usage-model change: daily prompt limits are gone, replaced by a “compute-used” model that factors prompt complexity + features used + chat length, refreshing every five hours up to a weekly cap. Closer to the OpenAI ChatGPT Plus pattern than the previous strict daily allotments.

For users currently on Google AI Pro at $19.99 who occasionally bump into limits, the new $100 Ultra tier is a meaningful intermediate option. For users on the previous $249.99 tier, the automatic $50/month price cut is a real consumer-friendly move.

What else shipped

Several substantial updates that are individually smaller but collectively meaningful:

Search and Shopping: AI Mode now powered by Gemini 3.5 Flash; new “Intelligent Search Box” with dynamic expansion; Information Agents that monitor the web 24/7 for updates (summer 2026, Pro/Ultra); Universal Cart unified shopping hub finding deals and price drops across vendors (summer, US).

Workspace AI deepening:

• Gmail Live — conversational email search (summer, Pro/Ultra, US)
• Docs Live — create and edit documents conversationally (summer, Pro/Ultra, global)
• Google Keep — AI-powered note organization (summer, Pro/Ultra)
• Google Pics — new AI image generation and design app

YouTube: Ask YouTube handles complex queries with structured video recommendations (Premium subscribers, US).

Android XR audio glasses — confirmed launch fall 2026:

• Hardware partners: Samsung / Qualcomm
• Design partners: Warby Parker, Gentle Monster
• Compatible with both Android phones and iPhone
• “Intelligent eyewear” framing — these are display-free audio glasses, not the AR headset variant

Android Halo — new feature showing agent progress at the top of the phone screen, launching later 2026.

Gemini app redesign — “Neural Expressive Design” with fluid animations, pill-shaped prompt box, single ‘plus’ menu, haptic feedback. Rolling out today on Android, iOS, and web.

What it means for Pick Right readers tracking AI tools

For Gemini users: The Gemini 3.5 Flash upgrade lands today, automatic. Spark agent access requires the new $100 Ultra tier (or above). The pricing restructure is consumer-friendly — existing $249 users get the cut, new mid-tier users get a $100 option. Worth evaluating whether the new $100 Ultra justifies the upgrade if you’ve been hitting Pro limits.

For ChatGPT users: The competitive shape tightened. Gemini 3.5 Flash on benchmarks is now in the GPT-5.5 ballpark, and the new $100 Ultra tier sits between ChatGPT Plus ($20) and ChatGPT Pro ($200) — that’s a real gap OpenAI doesn’t currently fill. Watch for OpenAI’s response.

For Claude users: Anthropic’s enterprise lead (per the May 13 Ramp Index) is unchanged by today — Claude’s writing-quality differentiation and the SpaceX Colossus capacity unlock remain the structural story. Spark’s MCP adoption strengthens the Anthropic-originated protocol rather than threatening Claude directly.

For AI video creators on Runway, Veo 3, or Kling: The big question — Omni’s pricing — landed paid-only rather than free-tier. The casual-creator video math doesn’t shift as dramatically as the pre-event analysis suggested. Runway, Veo 3, and Kling all stay competitive on their existing axes; Omni adds another paid option rather than disrupting the free-tier landscape. The pre-keynote preview speculated on the free-tier playbook; that prediction was wrong.

For developers using Gemini API: The Flash-tier upgrade is the most directly impactful change. If you’re running high-volume workloads, the 4x output speed claim with comparable-quality outputs likely improves your unit economics meaningfully. Try it today.

The structural read

Google’s I/O 2026 was substantive product news, not the demo-heavy showcase the genre is known for. Five real shipments (Flash, Omni, Spark, AI Ultra pricing, XR glasses confirmation) versus the usual one or two. The pricing restructure especially signals confidence — the $250 → $200 cut and new $100 mid-tier suggest Google sees Gemini’s market share growing fast enough to justify pricing compression rather than maintenance pricing.

The competitive picture in late May 2026:

• Anthropic leads enterprise adoption (Ramp data), $30B run-rate; differentiates on writing quality and Claude Code maturity
• OpenAI invested $4B in Deployment Company services to close the enterprise gap; GPT-5.5 Instant default tightened the consumer reliability story
• Google delivered model + agent + ecosystem depth today; pricing now competitive at every tier; Workspace integration deepens
• Open-source inference continues growing on commodity hardware

For users picking a primary AI tool, no single tomorrow-change. For users tracking which platform is winning the broader AI market, Google’s I/O bought back ground that had been slipping toward Anthropic and OpenAI through Q1.

For broader context on the AI competitive landscape, see the Gemini review, the Claude vs ChatGPT comparison, and the pre-I/O preview for what was expected vs what shipped. Affected reviews will be refreshed over the next 48 hours.

Questions or corrections? Email Pick Right. Want the full list? See all news.