Alphabet’s flagship AI model remains stuck in testing. Gemini 3.5 Pro missed its promised June launch. June became July. Now insiders wonder when it arrives at all. Shares have slid 15% from their 2026 peak. The pressure builds.
Google introduced Gemini 3.5 Flash at its I/O conference in mid-May. Executives positioned the lighter model as a speed demon for agents and coding. They promised the heavier Pro sibling would follow one month later. That timeline collapsed. Bloomberg reported the delay stretches months. Engineers retrained the base model after fresh data produced disappointing coding results. No new date has surfaced.
Frustration runs deep inside Google. Ten current and former employees told Bloomberg the lag has become a source of tension among engineers, researchers and managers. Many fear OpenAI and Anthropic now ship superior models. “The delay has been a source of frustration for Google engineers, AI researchers and managers, many of whom are concerned the company risks losing an edge in the market as rivals Anthropic and OpenAI produce models that exceed Gemini’s capabilities,” the report stated.
Yet the business tells a different story. So far. Google Cloud revenue jumped 63% in the first quarter to $20 billion. The unit’s operating margin hit a record 32.9%. Backlog nearly doubled to $462 billion. Enterprise customers keep signing up. Yahoo Finance detailed the surge, driven by AI services. Alphabet projects capital spending as high as $190 billion this year. That bet assumes frontier models stay competitive.
The Agent-First Strategy Takes Shape
Flash already powers real products. It became the default model in the Gemini app and AI Mode in Search. The company integrated it into NotebookLM for more accurate research. TechCrunch captured the shift at launch. Google positioned the 3.5 family around autonomous agents rather than simple chat. Flash handles execution. Pro, when it arrives, should orchestrate.
Koray Kavukcuoglu, Google DeepMind’s chief technology officer, praised Flash’s balance. “3.5 Flash offers an incredible combination of quality and low latency. It outperforms our latest frontier model, 3.1 Pro, on nearly all the benchmarks,” he said, citing gains in coding, agentic tasks and multimodal reasoning. The model runs four times faster than some frontier systems and shows 12 times optimization in targeted workloads. Internal demos showed it building an operating system from scratch. Partners automated complex workflows with human pauses at key decisions.
Tulsee Doshi, a Google AI product leader, outlined the tandem approach. “3.5 Pro becomes your orchestrator, your planner, and then it actually can leverage Flash to be the various sub-agents. I think it really comes down to where do you really want that reasoning power, where you actually want that larger model that can really push on the reasoning side versus where you have tasks that really do merit good brute force tool use capabilities.” The vision sounds coherent. Delivery lags.
Coding remains the proving ground. Enterprise buyers pay for models that generate reliable software, debug pipelines and manage long-horizon projects. Google updated training data specifically to lift those skills. Results fell short of internal bars. The company scrapped progress on one version and restarted pretraining. That decision adds time. It also signals caution. Better to ship strong than ship first and regret it.
But rivals refuse to wait. OpenAI’s o3 series and Anthropic’s Claude 4 models have posted strong gains in developer benchmarks this year. Some X posts from AI researchers claim certain Chinese open-source models now challenge frontier performance on cost-adjusted metrics. Public chatter amplifies the sense of slippage. Alphabet stock dropped more than 4% on the Bloomberg news in recent sessions. It closed Friday at $346.77 after peaking near $409.
Multiple layers of review slow Google’s process. The company weaves AI into Search, Maps, YouTube and Android. Each product team weighs in. Safety checks multiply. That structure protects brand trust. It also creates friction when speed matters most. Bloomberg’s sources highlighted this dynamic as a contributing factor beyond the retraining itself.
Analysts split on implications. Some see one delayed model as noise in a fast-moving field. Google still ships across dozens of variants and keeps costs low for customers. Its statement struck a measured tone. “We’re currently testing 3.5 Pro, an upgraded Flash model, and other models with partners. We are shipping quickly across a wide range of models while keeping them highly cost-effective for customers.” The words emphasize breadth over single-hero releases.
Yet investors watch the frontier closely. Capital expenditure at $190 billion demands returns. Google Cloud’s 63% growth and massive backlog suggest demand outstrips current capacity in places. Newer models could unlock larger deals. A prolonged gap might push some buyers toward Microsoft-backed OpenAI or Amazon’s growing AI offerings.
Recent coverage reinforces the tension. The Verge noted the missed June target and pointed back to Bloomberg’s sourcing on coding shortfalls. TechCrunch’s May coverage captured optimism around agents that now feels tempered by summer developments. No major new benchmark scores have emerged for an updated Pro. That vacuum fuels speculation.
Alphabet reports second-quarter results this week. Wall Street expects continued cloud momentum near 60% growth. Executives will face questions on timeline and capabilities. History shows Google rarely comments on unreleased models. The pattern holds. Silence amplifies uncertainty.
Inside DeepMind and Google Research, the reset carries weight. Restarting pretraining isn’t a minor patch. It resets the clock. Four senior researchers have reportedly departed in recent months, per social media discussion. Talent flight adds risk in a competitive hiring market.
Still, Google’s infrastructure advantage endures. Its TPUs power efficient training at scale. The company claims strong gains in efficiency and cost for the 3.5 Flash line. Those gains matter for production deployment even if raw benchmark leadership slips temporarily. Enterprises value reliability, integration and price more than leaderboard position in many cases.
The coming months will test that thesis. If 3.5 Pro emerges with meaningful coding gains and agent orchestration that exceeds current rivals, the delay becomes footnote. If competitors widen the gap further, investor patience may thin. For now the stock reflects doubt. Revenue growth tells resilience. Both signals matter. Neither is definitive.
Google built its empire on patient, iterative progress. Search took years to dominate. Android overcame early skepticism. The AI chapter follows similar contours. Flash delivers immediate value today. Pro’s arrival, whenever it lands, must justify the wait. The market has already priced in some disappointment. Further slips would test conviction more severely.
WebProNews is an iEntry Publication