Google shipped its second new Flash model in three weeks while the flagship everyone is waiting for remains missing in action

Google has shipped two new Flash-tier AI models in three weeks while its flagship model has slipped past a promised June launch and a mid-July date and remains nowhere in sight. The latest, Gemini 3.7 Flash, arrived August 13 with improved coding benchmarks at half the price of its predecessor 1. But the signal is in what Google did not ship: Gemini 3.5 Pro, the model that was supposed to be Google's next big answer to advanced models such as Anthropic's Mythos and OpenAI's GPT 5.6, was unveiled at I/O in May with a promise to launch the following month and has never appeared 2. Flash cycles now turn in three weeks; the flagship has been stuck since May.

What Flash is actually gaining

The new model posts measurable jumps over the 3.6 Flash it replaces, according to Senior Director Tulsee Doshi 1:

• FrontierCode 1.1 Main: 34.4% to 43.6%, a 26.7% relative gain
• DeepSWE v1.1: 49% to 65.3%, a 33.3% relative gain
• GDP.pdf, complex document processing: 22% to 34%
• AutomationBench, business workflows: 17% to 30.4%
• WebDev Arena: 1,538 to 1,588

The pricing is aggressive: $0.75 per million input tokens and $3.75 per million output through the end of the year, an introductory rate that halves what 3.6 Flash costs 1. Even halved, though, that price remains higher than what some rivals charge. The Flash-like Luna version of OpenAI's GPT 5.6 runs $0.20 per million input tokens and $1.20 per million output, which puts Gemini 3.7 Flash at 3.75 times Luna's input price and 3.1 times its output price 1. The rollout is narrower than the announcement reads, too: Gemini 3.7 Flash is live in the Gemini API, AI Studio, and Gemini Enterprise, but individuals only reach it inside the Gemini Spark agent with an AI Pro or Ultra subscription, and Google's regular chatbot interface still runs 3.6 Flash 1.

The flagship that never shipped

Google unveiled Gemini 3.5 Pro at I/O in May 2026 with a promise to ship it in June 2. It missed June. In late June, Google updated the data used to train Gemini to sharpen its coding skills, but the results were disappointing, one person familiar with the matter told the Los Angeles Times 3. It then missed a mid-July date 2. Fortune counts three missed release deadlines in recent months but names only these two. Google headed into August with 3.5 Pro still unreleased, even as it claimed to have begun training Gemini 4 2.

The internal picture explains the stall. The Los Angeles Times describes Google Cloud, Google DeepMind, and the Android team all building overlapping coding tools, with some engineers resisting AI-generated code on quality grounds 3. A Google spokesperson framed the wider strategy this way: "We're shipping quickly across a wide range of models while keeping them highly cost-effective for customers" 3. The week before 3.7 Flash shipped, DeepMind cofounder Demis Hassabis gave up the CEO title to become chairman, and day-to-day operations went to DeepMind's chief technology officer Koray Kavukcuoglu, reporting directly to Sundar Pichai from Mountain View rather than London 2. Fortune reports that Google's models are "currently not competitive with the bleeding edge of the U.S. frontier on intelligence and coding benchmarks" 2. The Fortune report ran August 10; Gemini 3.7 Flash shipped three days later, on August 13 1.

What the release cadence hides

Read the Flash releases against the Pro delay and a pattern takes shape. Google can iterate rapidly on a mid-tier model, post double-digit benchmark gains, and cut prices to hold developer attention. What it apparently cannot do is produce a flagship that clears the bar set by rivals. Micah Hill-Smith, cofounder and CEO of independent benchmarking company Artificial Analysis, told Fortune that Gemini 3.6 Flash, the strongest model Google has shipped, now sits behind models from Anthropic, OpenAI, one or two leading Chinese labs, xAI, and Meta on his firm's intelligence index 2. Three DeepMind engineers told Fortune they blamed the delays on Google failing to prioritize AI coding abilities 2. Yet Google said at its most recent Cloud conference that 75% of code at the company is now generated by AI 3.

For builders choosing an API today, the calculus is concrete: Gemini 3.7 Flash is incrementally better than 3.6 Flash at half the introductory cost, and still 3.75 times OpenAI's Luna on input price. For investors, the question is whether Google's strategy has shifted from a defining frontier model to release cadence as the competitive position, velocity in place of a flagship moment it cannot yet win. The Flash models are real, the improvements are measurable, and the price cuts are genuine. But they are shipping in the exact window when the model Google needs to compete at the frontier slipped past its June and mid-July dates.