Google is launching three new variants of its multimodal large language model (LLM) Gemini. The new Gemini 3.6 Flash should not only deliver better inferences, but also use around a sixth fewer tokens than its predecessor, Gemini 3.5 Flash. At the same time, the tokens themselves become cheaper.
Read more after the ad
But Gemini 3.5 is far from obsolete. Rather, it has two siblings: 3.5 Flash Lite is said to have lower latencies than the 3.5 Flash presented in May at the Google I/O developer conference. Google promotes Flash Lite specifically for AI agents.
Where is Pro?
What is still missing is the Pro version of Gemini 3.5, which works more precisely and is intended for more complex tasks. According to a Bloomberg report, the data company is plagued by quality problems. According to insiders, the data company is also integrating its AI into many of its own products at the same time, which can be as different as Google Maps and YouTube. This would mean that many Google projects would have a say, which wouldn’t exactly speed things up.
Google itself said in its announcement on Tuesday that it is testing Gemini 3.5 Pro with partners and plans to make it generally available “as soon as it is ready.” Google would rather talk about the preparations for training the upcoming Gemini 4.
Flash Cyber
The US company wants to keep IT security enthusiasts happy and in line with Gemini 3.5 Flash Cyber. Flash Cyber is designed to work with the in-house AI agent Codemender and, as the name suggests, specializes in checking software for security vulnerabilities.
Read more after the ad
However, only government services and some institutions selected by Google are allowed to use Flash Cyber. Codemender can also integrate other major language models in parallel.
(ds)
