Google DeepMind Strategic Pivot: Shelving Gemini 3.5 Pro to Focus Compute on Gemini 4
Following the recent leadership reorganization under SVP Koray Kavukcuoglu, Google DeepMind has executed a sweeping strategic pivot in its foundation model roadmap. The laboratory has quietly shelved its long-delayed flagship Gemini 3.5 Pro model—originally scheduled for a June 2026 release—in order to channel all available pre-training compute and engineering resources directly into its next-generation frontier architecture, Gemini 4 (specifically Gemini 4 Pro).
To maintain market presence and combat competitive pressure from OpenAI's newly released GPT-6 Astra and Anthropic's Claude Fable, Google DeepMind is pursuing a dual-track strategy: a rapid-fire release of highly efficient, low-cost "Flash" models to satisfy immediate commercial developer demand, while reserving its core supercomputing clusters to train the Gemini 4 flagship, which is targeted for an October 2026 launch.
Kavukcuoglu's New Model Strategy
- Shelving Gemini 3.5 Pro: The decision to cancel Gemini 3.5 Pro represents a pragmatic acknowledgement of training constraints. Rather than deploying a model that was delayed due to tool-calling failures and risked being immediately outpaced by competitors, DeepMind has redirected its resources to ensure Gemini 4 Pro can compete directly with GPT-6 Astra.
- The First Gemini 4 Pro Checkpoint: Google has officially deployed its first internal pre-training checkpoint for Gemini 4 Pro, signaling that the model has progressed past early-stage training, though it continues to face pre-training scalability challenges.
- The "Flash-Lite" and Incremental Wave: While the flagship Pro tier remains sidelined, DeepMind has launched a series of incremental models at a breakneck pace. These include Gemini 3.8 Flash (internally code-named "Skimaki") and Gemini 3.8 Flash Cyber, both released on September 2, 2026. These models are designed to deliver high-speed, sub-second agentic loops and cost-competitive coding capabilities (matching Anthropic's Claude Opus on key benchmarks at a fraction of the cost).
This shift marks a clear signature of Kavukcuoglu's day-to-day operational leadership: prioritizing agility, aggressive pricing, and a focused consolidation of compute on a single, clean frontier breakthrough rather than incremental flagship releases.