# Google Has Silently Canceled Gemini 3.5 Pro, Says Semi Analysis Report > Source: > Published: 2026-08-10 08:44:31+00:00 Google’s much-awaited Gemini 3.5 model might never release at all. That’s the claim from a new institutional note by Semi Analysis, which argues that Google has quietly shelved Gemini 3.5 Pro and is now pivoting to hype around Gemini 4 to paper over the gap. The report is blunt about where this leaves Google in the model rankings, and it isn’t a flattering place. The saga began at [Google IO in May](https://officechai.com/ai/gemini-flash-3-5-benchmarks/), where Sundar Pichai stood on stage and told a room full of developers that Gemini 3.5 Pro was coming. Google released Gemini 3.5 Flash that day instead, a lighter model built for agentic and coding work, and Pichai asked the crowd for patience on the bigger model, saying it would arrive “next month.” The line drew an audible groan from the audience, according to multiple reports from the event, and set a deadline that Google would go on to miss badly. June came and went with no Pro model. Instead of the flagship, Google spent July retraining Gemini on updated data specifically to fix coding performance, and when that work fell short of internal expectations, the company shipped [Gemini 3.6 Flash](https://officechai.com/ai/google-releases-gemini-flash-3-6-and-gemini-flash-3-5-lite/) on July 21 as a stopgap alongside Gemini 3.5 Flash-Lite and a cybersecurity-focused Flash Cyber variant. Gemini 3.6 Flash improved on efficiency, using around 17 percent fewer output tokens than its predecessor while beating it on benchmarks like DeepSWE, OSWorld-Verified, and MLE-Bench. What it didn’t do was move the needle on raw intelligence. Independent testing from Artificial Analysis gave Gemini 3.6 Flash the same Intelligence Index score as Gemini 3.5 Flash, meaning the model everyone was told would tide them over turned out to be a lateral shift rather than an upgrade. Rivals noticed the gap too, with [OpenAI and Meta staffers appearing to mock Google](https://officechai.com/ai/openai-meta-execs-appear-to-mock-google-after-release-of-gemini-flash-3-6/) on X the same day the models dropped. The Semi Analysis note points out that this positions Gemini well behind the frontier. Gemini currently sits in eighth placed lab in the Artificial Analysis Intelligence Index, trailing Muse Spark 1.2, Grok 4.5, and the top tier of Chinese open-source labs, all of which the report says now outperform Gemini 3.6 Flash on a consistent basis. For a company that broke through a $4 trillion valuation partly on the strength of Gemini 3’s reception late last year, slipping outside the top five is a rough turn. There have been signs of life. Gemini 3.5 Pro has surfaced on Arena a handful of times over the past few weeks, briefly appearing in blind testing pools before being pulled, in one case after less than an hour. Each sighting has set off a fresh round of speculation on X that a launch is imminent. But the reaction to the outputs themselves has not matched the hype. Testers who ran their usual prompts against the model reported responses that felt incremental rather than the generational leap Google had been teasing since May, with one widely shared thread describing the outputs as adding a lot of surface-level features while lacking any real detail underneath, calling the results a lazy skeleton dressed up to look finished. Then came the detail that could be the real signal. Last month, alongside the Gemini 3.6 Flash launch, Google said it had already begun what it called its “most ambitious pre-training run yet” for [Gemini 4](https://officechai.com/ai/gemini-3-5-pro-is-testing-with-partners-have-begun-pre-training-run-for-gemini-4-pro-google/). Frontier pre-training runs take months of compute and iteration before a model is even ready for internal testing, which means Google was working on Gemini 4 while Gemini 3.5 Pro was still supposedly weeks from release. That timeline is difficult to square with a company that’s actually planning to ship the model it announced in May. It reads more like a team that has moved on and is hoping nobody asks too many follow-up questions about the one that got away. Semi Analysis isn’t optimistic that Gemini 4 fixes any of this either. The report argues that whatever structural issues held back Gemini 3.5 Pro, particularly around coding and agentic reliability, aren’t the kind of problems a bigger pre-training run automatically resolves. Google has been open about lagging behind on agentic coding specifically, going as far as forming a dedicated DeepMind coding team to close the gap. Until that shows up in a released model, the report’s ranking stands: Gemini in eighth place, with a canceled flagship and a promise of something better that keeps getting pushed further out.