Google delays Gemini 3.5 Pro to rebuild its base model while Meta ships cut-price Muse Spark 1.1, two rational bets on where 2026 enterprise AI budgets actually go.
Two frontier labs made opposite moves this fortnight. Google DeepMind pushed the launch of Gemini 3.5 Pro to July 17, confirming it scrapped the 2.5 Pro-derived architecture entirely and rebuilt the base model after enterprise testers flagged coding gaps, token-efficiency issues and long-task reasoning short of flagship standards. The rebuilt model is said to target stronger mathematical reasoning, a 2-million-token context window and a "Deep Think" reasoning layer. The flagship had been promised for June; at I/O in May, Sundar Pichai asked the audience to "give us until next month."
Meta, meanwhile, went down-market and fast. On July 9, Meta Superintelligence Labs released Muse Spark 1.1, a multimodal reasoning model built for agentic and coding work, with a self-managed 1-million-token context window, native primary-agent and sub-agent orchestration, and MCP and custom-skill support. Pricing is the story: $1.25 per million input tokens and $4.25 output, above budget tiers like Claude Haiku 4.5, but far below the flagships (Opus 4.8 at $5/$25; GPT-5.5 at roughly $5/$30). AI chief Alexandr Wang called it Meta's "strongest model for agentic and coding work yet." The preview is US-only at launch.
Our Take
Google is defending flagship credibility at the cost of cadence; Meta is conceding the frontier crown to attack the volume market just as enterprise AI spending comes under CFO scrutiny. Both are rational responses to the same fact, the middle of the market, not the peak, is where 2026 enterprise budgets are actually being spent. Watch whether Gemini 3.5 Pro's July 17 arrival justifies the rebuild; a second slip would be materially damaging.