Google rolls out agentic Gemini Spark and pitches Gemini 3.5 Flash for cost control

Google began rolling out Gemini Spark, an agentic assistant that can execute digital errands autonomously — even when a user's phone is off — to AI Ultra subscribers in the US following Google I/O 2026. It represents Google's clearest move yet from chatbot to autonomous agent, joining the broader assistant-to-agent pivot also seen in OpenAI's Codex and Microsoft's async Copilots.

The more strategically pointed announcement was Gemini 3.5 Flash, which Google positions as rivaling frontier models while sharply cutting token costs. CEO Sundar Pichai explicitly warned that companies are 'blowing through' their annual token budgets — a pitch aimed squarely at the cost-fatigue gripping enterprises. Google also shipped Gemini Omni, a Gemini UI redesign, and a macOS desktop app, with Sundar and Josh Woodward acknowledging user feedback about hitting rate limits too quickly and rolling out fixes.

Mechanically, Flash's value proposition is intelligence-per-dollar parity with bigger models for high-volume workloads, directly echoing DeepSeek's and Anthropic's price moves. The competitive framing is explicit: Google wants to be the rational choice for teams whose AI bills have spiraled.

Skeptics weren't sold. CNET's 'Mystery Meat' critique of Gemini and debates over whether redesigns deliver real utility show users questioning substance behind the launches. Separately, Google's confirmed Gemini 3.5 Pro is slated for June to close a reasoning gap the Flash line had regressed on. Watch whether Spark's autonomous errands prove reliable (and safe) in the wild, and whether Flash's cost pitch actually shifts enterprise spend away from rivals.