Google announced the start of Gemini 4's pretraining on July 21, 2026, describing it as the largest pretraining effort in the company's history. As of early September, the model has produced strong preliminary results from that phase. However, posttraining has not yet begun — and that is the stage where AI models are transformed from impressive raw computations into products people can actually use. Posttraining — which includes fine-tuning, reinforcement learning from human feedback, and instruction tuning — is widely credited with shaping how reliably a model follows instructions, handles tool use, and behaves safely in deployment. It is often where release timelines slip, which is why its progress matters as much as pretraining results.

What Gemini 4 is being built to do

CEO Sundar Pichai outlined the ambition during Alphabet's Q2 earnings call on July 23, framing Gemini 4 around two specific priorities: coding and autonomous agents. Pichai also said Gemini 4 would rely on what he called "much larger base models" to meet the demands ahead. Both priorities align with where enterprise demand for AI has concentrated: coding assistants have become one of the fastest-adopted AI product categories among developers, while agentic systems — models that can plan and execute multi-step tasks — are the focal point of competition among the major AI labs.

The spending reflects that scale. Alphabet has set its 2026 capital expenditure guidance as high as $205 billion, a figure that has pushed the company into negative free cash flow territory for the first time.

Where the competitive pressure comes from

Gemini 3.5 Pro, the model that preceded this effort, reportedly experienced delays tied to coding issues. Gemini 4 is, in part, a response to that. A model explicitly built around coding strength addresses one of the specific areas where competitors — most notably OpenAI and Anthropic, whose models have been strong performers on coding benchmarks and developer adoption — have landed their most credible blows against Google's prior releases.

Analysts tracking Google's historical training cycles view a late-2026 release window as the most plausible scenario, roughly six months after pretraining began in July. The company has not publicly disclosed a formal timeline.

What comes next, and why posttraining is the variable to watch

As of September 1, 2026, no public benchmarks for Gemini 4's pretraining or posttraining timelines have been released. For a company that spent much of 2025 and early 2026 rebuilding developer confidence in the Gemini product line, the silence reads less like strategic secrecy and more like disciplined expectation management. Watch for early signals such as public benchmark submissions, developer previews, or integration into Google's own products — the channels through which past Gemini releases have typically surfaced first.

The broader tech sector has been watching Alphabet's capital spending trajectory closely, partly because it signals how much incumbent players believe the current AI investment cycle still has runway. A $205 billion capex year from a single company has ripple effects across chip suppliers, data center construction, energy infrastructure, and the startups competing for the same GPU capacity.