tl;dr

• Is Gemini 3.5 Flash the best budget model?
• Gemini Omni video editing model

Releases

Gemini 3.5 Flash

GoogleIO was this week, where they released dozens of new (mostly AI) products and features, with the highlight being Gemini 3.5 Flash.

Google’s previous models have struggled in the real world, with the most recent Gemini 3.1 Pro model not even being that much better than its predecessor. This pattern looks like it is continueing for Gemini 3.5 Flash as well.

Despite the strong benchmark scores, there are a plethora of people online who are unhappy with how this model performs for real world tasks. People have reported it struggling with basic math problems (does 300 + 140 = 360?), producing large amounts of slop code that doesn’t work, and performing worse with additional guidelines that are meant to help it perform better, which may be due to the fact that it seems scared of its own system prompt.

This could be forgivable, as the Gemini Flash models are meant to be the budget option that don’t necessarily give you the best performance, but are a reasonable price. The issue is that despite being built from Gemini 3 Flash, Google has decided to increase the price by 3x, costing $9 per million output tokens. For reference when GPT 5 was released, it was only $1 more expensive (GPT 5.5 is now $30 per million output).

Not only does the model cost more, but it uses more tokens to answer questions as well. Due to these two factors it costs 2x more than Gemini 3.1 Pro to run the Artificial Analysis benchmark suite.

Cost wise this puts the model around the same price as GPT 5.5 with medium reasoning, and to add insult to injury, despite Gemini Flash 3.5 being a benchmaxxed model, it still performs worse that GPT 5.5 medium and is slower due to generating so many more tokens.

The budget model is not cheap, its speed is being used to go nowhere very quickly, and it smarts are on paper only.

Stay away from Gemini 3.5 Flash.

Quick Hits

Gemini Omni

Unlike the LLM team, the image and video generation teams at Google are doing much better, as seen by their new Gemini Omni model.

It is meant as a video editing model primarily, and is the first step of Google’s foray into a truely Omni-modal model that can make anything.

There’s no 3rd party benchmarks for it yet (we dont have any good video to video leaderboards in general), but based on their strong Veo 3.1 model, I expect this model to also be very strong as well.

Finish

I hope you enjoyed the news this week. If you want to get the news every week, be sure to join our mailing list below.