Google just dropped Gemini 3.7 Flash with massive speed improvements and impressive coding benchmarks. We compare it to Claude and GPT to see if Google’s new workhorse model can finally dominate the AI race.
Full transcript
Today, we're going to be talking about Google and Gemini 3.7 flash that just dropped, only three weeks after 3.6 flash, and this is basically like their workhorse flash model, so it's designed for speed, it's designed for intelligence on coding, agentic tasks, we've got Casra Dash here as well. Yo, yo, yo. So, I'm looking at the numbers now, and they've actually done a massive job, have you got the numbers in front of you? Should I share them?
Yeah, pull them up. Alright, let's pull up. So, here is the actual test bench from 3.6 flash to where they are now. As you can see, and did you say it was a month from 3.6, or three weeks?
Yeah, only three weeks, yeah. So that's a massive increase, especially in the flash, like their quick model. So they've went from 49.53, 17 to 30.4, and 34.46, and in comparison, Claude, it's actually relatively cheap. Claude Fable is $10 per, this is $3.75.
Yeah, that's a lot cheaper, isn't it? I mean, it's basically like their lighter model. So they've got Gemini 3.5 Pro, that should be coming out, and everyone's like, well, where is it? You know, where's their flagship model?
But this one, yeah, they focus on speed, and it's designed to be like super fast. So if you have a look, for example, it's pretty much up there in terms of production code quality on front tier code 1.1. So if you look at Gemini 3.7 flash versus Claude Sonnet, it's actually beating Claude Sonnet 5, which is pretty impressive in itself, and GPT-5.6 Terra. I've actually noticed a little bit of you more.
Yes, yeah, I feel like Claude's quality really varies, particularly Sonnet 5. It gets distracted, it's slow, it's nowhere near as good as Fable 5, even though on benchmarks it should be. There's something like qualitative to models that there's no benchmark for yet. It's like almost like focus or getting stuff done.
Yeah, I also think they deteriorate over time as well, because I remember like 3-week log, or even like when I've tested out, it's been really good, and then it's slow, dumber. Yeah, yeah, I would agree. Like it's getting worse and worse over time. But yeah, it's kind of a model where it's designed for speed, but I feel like who's going to be using this apart from Google themselves?
Like it's probably great for them on their search results and providing faster search results. That's exactly what I was going to say. I was going to say, I think this is probably going to be AI mode results. So not even the AI overview, but for AI mode where people are going into the AI mode tab search, like how I couldn't see anyone coding with this.
I couldn't logical thinking with this. Out of curiosity, do you use any of the Google LLMs for it? No, I don't find myself using them at all. Like, for example, for the images, I'm going to use GPT image 2 for coding.
I'll use Claw. For agents, you know, we might use something like DeepSeek, but I don't see anywhere where Google naturally fits into the ecosystem. Like even for video, you're probably going to use something like Grok for B-roll or something like that. Minimax.
SeedDance. Yeah, SeedDance. HiggsField. Yeah, it's a bit of a difficult because like I feel like there's no ecosystem around Google's anything.
Like obviously they've got Google search, they've got the Google AI overviews, but to my knowledge, they don't have like a Codex or a Claude code, do they? They've got Gemini Spark. So Gemini Spark is kind of like their all-in-one desktop app, but I don't know who they're marketing to. It's insane, according to some YouTube channels out there.
Honestly, it's so nice. It's one of those where there's just like so many different options. You just pick one of the best and I think Claude is the best. Claude is the all-out.
Yeah, like there's, I mean, apart from the AI overviews and obviously as SEOs, we always want to try and get site overviews, getting our brand and name meant. I do not see Gemini winning iRace, which is very scary to say that because a year ago, if you had asked me, I would have said I reckon Google's most amount of data, they've indexed and crawled pretty much the entire web and they've got the history of doing that. But now, yeah, hopefully the mix-up that we spoke about a few days ago of the actual running order will change that. No, I don't.
I don't know many people that use it and everything's just kind of right in the middle. Everything's kind of mediocre. So they really have to step up. And then a lot of their team has either stepped down or stepped up or moved around.
Like they had a big shake-up last week where, you know, one of the CEOs left. So crazy situation. I think all their best talent is leaving as well. Yeah, it's going to be an interest.
I currently, though, maybe more like I reckon it's probably Claude, right? Then it's maybe Chad, maybe one of the LLM. Yeah, they've got to bring something big when they release Gemini 3.5 Pro. It has to be something impressive because that's what everyone is waiting for.
We will see. And obviously, we'll probably do an update on that. Yeah. Thanks for watching.
Link in the comments description if you want to learn SEO with Kasra's Rank OS. And if you want to automate your business, check out the Airproft reporting. Link in the comments. Thanks for watching.
More episodes