Google’s Gemini 3.5 Pro might’ve been delayed, but it appears that it could launch a new Flash model in the coming days.
A developer poking around Google’s Antigravity coding platform spotted references to a model called “gemini-3.6-flash-tiered” in the app’s configuration files, and posted screenshots of the find on X. The listing includes a full spec sheet: image support, thinking mode enabled by default with a thinking budget that can be set to unlimited, a context window of over a million tokens, and a max output of 65,536 tokens. It’s also been slotted into Antigravity’s tiered model system as the new “flash” tier, sitting above “flashLite” and below “pro” in the naming structure Google uses internally.
None of this has been confirmed by Google, and there’s a real chance the model doesn’t ship under this exact name or timeline. But the fact that it’s already wired into Antigravity’s backend, complete with quota tracking and a reset timer, suggests it’s further along than a rumor. Google has a habit of testing new models quietly inside its own developer tools before making any public announcement, and that’s exactly the pattern that played out with Gemini 3.5 Flash back in May.
The timing matters here. Gemini 3.5 Pro was supposed to arrive in June, according to what Google said on stage at I/O. That deadline came and went, and Bloomberg reported the delay is tied to the model falling short on coding benchmarks internally, with Google reportedly retraining on updated data in late June only to see disappointing results again. Google has confirmed it’s currently testing 3.5 Pro with select partners and the US government, but there’s still no public release date. For a company that has talked openly about wanting to close the gap with OpenAI and Anthropic on agentic coding, shipping a flagship model that underperforms on exactly that metric is not a great look.
A Flash release would give Google something to point to while Pro keeps cooking. It’s a move the company has used before. When Gemini 3.5 Pro missed its original window, Google leaned on Gemini 3.5 Flash instead, and that model ended up beating Gemini 3.1 Pro on several coding and agentic benchmarks despite being the cheaper, faster tier. If 3.6 Flash follows the same script, it wouldn’t need to compete with GPT-5.6 or Claude Fable 5 at the very top of the leaderboard. It would just need to keep Google’s name in the conversation.
That conversation has gotten a lot more crowded lately. Google recently fell out of the top five labs on the Artificial Analysis Intelligence Index for the first time, with Gemini 3.5 Flash sitting at a score of 50 while Anthropic, OpenAI, SpaceXAI, Meta, and even an open-source Chinese lab all had models ranked higher. The gap has only widened since. In the space of about a week, SpaceXAI’s Grok 4.5, three separate GPT-5.6 variants from OpenAI, Meta’s Muse Spark 1.1, and Moonshot AI’s Kimi K3 all launched, pushing the number of labs with a model scoring above 50 on the index from two to six. Claude Fable 5 still holds the top spot, but the models chasing it now come from five different companies, not two.
Google isn’t sitting still, obviously, but it is sitting behind, and that’s the backdrop against which any Gemini 3.6 launch will be read. A well-timed Flash model won’t fix the Pro situation, and it won’t undo months of delay. What it can do is remind developers that Google is still shipping, even if the model everyone’s actually waiting for hasn’t shown up yet. Whether “gemini-3.6-flash-tiered” turns into a real public release, or just another internal build that never sees daylight, should become clear soon enough given the pace at which Google has been testing things inside Antigravity lately.