Google debuted Gemini 3.8 Flash on Wednesday, the fourth Flash-branded model the company has released since May and the third in just over three weeks. The company says the model matches larger rivals' performance on some benchmarks at a much lower cost, and that it excels at coding.
Flash is Google's label for its smallest, fastest, and cheapest models, aimed at users who want speed without giving up too much capability. But Gemini 3.5 Pro, the flagship model CEO Sundar Pichai said would ship in June, remains unreleased. It was still in testing in July, and Google's own website still lists it as 'coming soon.' According to the Wall Street Journal, internal candidates for the Pro model were scrapped because they failed to sufficiently outperform Flash.
The delay has not gone unnoticed by competitors. Meta chief AI officer Alexander Wang mocked Google on X this week — 'I really hate to say it, but…gemini who?' — after his own company's Muse Spark 1.3 model leapfrogged Google's lineup on the Artificial Analysis Intelligence Index. On that benchmark, Google's best model, Gemini 3.8 Flash, now ranks 10th.
Google is framing its Flash cadence as evidence it leads on 'recursive self-improvement,' or RSI, a technique in which AI models help refine their own successors. In its release announcement, Google said Gemini 3.8 was 'further accelerated' by long-running AI-agent loops that 'recursively evaluate and refine the underlying models' — a more direct claim than in prior Flash rollouts, which described agents building games or training a robotics model.
The economics explain part of the strategy. Flash models require less computing capacity to modify, according to people cited by the Journal, letting multiple research teams test approaches in parallel; changing the larger Pro models demands far more resources. Pichai has called Flash the company's 'workhorse' series, hitting the 'sweet spot of performance and cost.' On its Q2 earnings call, Google said its model APIs were processing about 22 billion tokens per minute, up from 16 billion the prior quarter, while computing supply remained constrained.
That combination — surging demand and limited compute — gives Google a clear commercial reason to keep polishing Flash rather than divert scarce capacity to Pro. Whether the company can still leapfrog Anthropic's Mythos 5 or OpenAI's Astra with its flagship model remains an open question.
The numbers come first, and they tell a simple story: Google is optimizing for the product it can ship profitably today, not the one investors were promised months ago. Capital rewards clear rules and delivered results, not roadmaps. Every quarter that Gemini Pro slips is a quarter rivals get to define the frontier, and a market that runs on compute scarcity and margins will not wait for a press release.
For a company that built its dominance on being first, the lesson here is unforgiving: in a free market for intelligence, speed to shipment is itself a form of discipline, and Google's own workhorse strategy may be the clearest signal yet that its moonshot has stalled.
