Google confirms Gemini 4 in training and expands Gemini Spark agent access

Google is teeing up its next frontier model: Sundar Pichai confirmed Gemini 4 is in training with much larger base models, with a likely launch in November or December 2026. In parallel, Google is expanding access to Gemini Spark, its agentic AI assistant, to more AI Pro and Ultra subscribers — pushing agentic capabilities into its consumer tiers.

The capital backdrop is enormous: Alphabet's 2026 capex is projected at $195–205 billion, overwhelmingly for data centers and custom silicon — consistent with the week's compute-investment theme and Google's server-chip/Gemini-in-hardware ambitions. The scale of spend signals Google intends to compete at the frontier through infrastructure as much as model architecture.

This is a continuation of Google's Gemini roadmap with a specific new fact — the confirmed Gemini 4 training and late-2026 window — rather than a re-announcement. It also lands against a backdrop of internal challenges: a Next Web report tied a Gemini delay to employee morale and talent-retention problems at DeepMind, and Meta's Alexandr Wang trolled with a 'Gemini who?' post. The competitive question is whether Gemini 4 arrives on schedule and delivers a genuine capability jump, or whether Google's near-term story remains efficiency-and-cost (Flash models, custom chips) while rivals ship frontier updates faster. Watch the Q4 launch window and any independent Gemini 4 benchmarks.