Google has announced the release of three distinct new models for its Gemini AI tool, one designated as 3.6 and two as 3.5. And while the division of labor for each model is detailed in the announcement, some other things stand out. One is the very early mention of Gemini 4. But one or two other queries have raised a few awkward questions from users about Google’s Gemini roadmap.

Gemini 4 reveal

The three new models of Google’s AI tool announced on July 21 were Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash-Cyber in CodeMender. If you’re not sure about each, here’s the summary. 3.6 Flash is the new multimodal workhorse for Gemini. 3.5 Flash-lite is the fastest and most cost-effective model of Gemini. Lastly, 3.5 Flash Cyber in CodeMender is a specialist cybersecurity-focused model for enterprises. The announcement focuses on those three, but it also offers a sneak peek of the future.

Beyond today’s releases, Gemini 3.5 Pro is currently testing with partners and we plan to make it broadly available as soon as it’s ready. In parallel, our team is already focusing on building the next generation of models. We have started our most ambitious pre-training run yet, for Gemini 4, and are excited by the progress.

blog.google

Awkward Gemini questions

What’s obvious is that Gemini 4 is a little way away, but it was one of the things seized upon by users in a thread on the Gemini SubReddit. The second most upvoted comment? “3.5 Pro flop confirmed”, with one supporting comment noting “Fr [for real]. Like, how bad must it be if are they [sic] pre-hyping the [Gemini 4] model after their unreleased [Gemini3.5 Pro] model?”

That’s not the only question, though, with one user focusing on Google’s ‘3.5 Flash-Cyber in CodeMender’ model, asking: “I don’t get it why they have to have a competing model with Anthropic. Code Assistance is not Google’s business. Never was. It would make more sense for them to either focus on releasing reliable yet resource efficient models. Or to work on technology to improve efficiency of AI models in general.”

One reply to that suggests: “Because they want that subscription $ from the coding population. That simple.” Another user backs that view, with: “Because I pay 200€ + 250€ in extra usage to anthropic every month and nothing to gemini.” Meanwhile, a third user took a dig, stating that Google could “Reclaim the frontier for like 2 weeks then downgrade the model like they always do.”

Some bright sparks

However, the comments aren’t all negative towards the new Gemini model news. In light of the announcement, and the original post’s question of “Do you think Gemini 4 can close the gap with OpenAI and Anthropic?” one Redditor suggests, “Google can win on 2 fronts – super efficient low cost models like flash light with multimodal and super intelligent science based model (like Gemini 3 was)”.

One Redditor offered a more laid-back and detailed response to Gemini’s competition with the likes of Claude and ChatGPT, saying: “Google is different to OpenAI and Anthropic in the sense that their AI is outputted through dozens if not hundreds of services that were not designed with AI in mind”, adding:

“Im [sic] not surprised Google struggles to maintain a lead in frontier models: technically, they don’t have to. Their user base will wait for competent gemini models to appear inside the software they already pay for and use.” That’s a fair point. Point towards the money, or point toward the delayed Gemini 3.5 Pro. Either way, Gemini 4 is clearly in Google’s plans, and by then we may see exactly where it stands versus its competition.

About the Author

Kevin Pocock

Kevin has a broad interest and enthusiasm for consumer electronics, PCs, and all things consumer tech. And more than 17 years experience in tech journalism.