Google DeepMind has transitioned its next-generation model, Gemini 4, into the early post-training phase. Koray Kavukcuoglu, elevated to Senior Vice President of Google DeepMind on August 12, 2026, confirmed the milestone during his first public appearance at The Information AI Agenda Live Summit on September 23-24, 2026. This marks a critical pivot for a company whose current frontier offering sits 40 percent behind the market leaders on the standard intelligence benchmark.
Kavukcuoglu signaled an aggressive timeline. “Our intention is to roll out an early post-training version as soon as possible, because we’ve already seen promising results and are very excited,” he said. No specific release date was provided, but the target is much earlier than the end of the year. The pace is notable: Google announced the start of its most ambitious pre-training run for Gemini 4 on July 21, 2026. A transition from pre-training to post-training in approximately two months is a compressed timeline, particularly given that Gemini 3.5 Pro has remained stuck in partner testing since June 2026.
The Capability Gap
The urgency is legible in the numbers. On the Intelligence Index v4.3.2, Google’s current frontier model, Gemini 3.6 Flash, scores 34.0. That trails Anthropic’s Opus 5.5 at 57.6, GPT-6 Astra at 52.7, and OpenAI’s GPT-6 Sol at 47.5. The 24-point gap between Gemini 3.6 Flash and Opus 5.5 is not a marginal deficit — it is the difference between competing at the frontier and competing below it. Gemini 4 is Google’s attempt to close that distance.
One Week, Four Labs, Zero Restraint
The timing of Kavukcuoglu’s announcement is the story. Between September 21 and September 22, the market absorbed Grok 4.7 at $2/$6, Opus 5.5 at $4/$20, and GPT-6 Sol and Luna at a permanent 50 percent price cut. The coordinated pricing structure that held through the Amodei pacing framework collapsed in under two days. By confirming Gemini 4’s post-training status immediately after this wave, Google is signaling it intends to compete in the next cycle — not watch from the sidelines.
But the details Google has not disclosed matter as much as the ones it has. No architecture. No parameter count. No pricing. No benchmarks. What is known is the model’s stated focus: coding, autonomous agents, and long-horizon agentic workflows — capabilities that move beyond text generation into active task execution. Google has said Gemini 4 is “significantly larger” than prior models. The question is whether larger translates to competitive on the benchmarks that now define the market.
The Evaluation Tension
The broader context sharpens the stakes. This week, the industry debated who gets to define safety evaluation — from the UN Security Council briefing where lab CEOs testified about their own models’ risks, to OpenAI publishing its own assessment framework. The push to ship faster sits uncomfortably alongside the demand for more rigorous evaluation. Google is compressing the timeline to market at the same moment the industry is asking whether post-training evaluation is thorough enough to catch the failures that have defined this quarter — from the Australian Medicare breach to the Hugging Face incident.
For builders evaluating model routing strategies, Gemini 4’s arrival — whenever it arrives — introduces a new variable into an already volatile stack decision. The current pricing war means cost-per-task is the new battleground. If Gemini 4 enters at frontier-class performance with competitive pricing, it could redistribute the developer ecosystem. If it enters below the bar, Google’s position in the AI agent economy weakens further.
Investors should watch two signals: whether the leadership reshuffle that elevated Kavukcuoglu over Gemini development translates into execution speed, and whether the compressed post-training timeline produces a model that meets or misses the capability threshold the market now expects. The three-way frontier race has not been this compressed since the GPT-4 era. Google has chosen to accelerate rather than wait. Whether that choice produces a competitive product or a premature one is the defining uncertainty of the next quarter.