m@growthschoOkay no seriously, I am writing this on a Thursday night and all I can see is drum rolls for Opus 5 and speculations that there’s quite rollout ongoing.

And it’s probable it comes out before the weekend hits, and you’re most likely reading this on a Saturday or a Sunday.

So, let’s see.

I wanted to talk about Google in the meanwhile.

I had noted down this topic idea on June 30 lol, hardly got a chance to cover it.

Since then, Google’s released a new set of flash models, and still have kept all of us hanging for their Pro / Gemini 4 / SOTA models, so much so that people are starting to lose hope.

And the memes on Twitter are too good.

But anyway I thought I’d just see it for myeslf. Surely it can’t be that bad, right?

But first, some catchup:

The big open model everyone's posting about, Kimi K3, can build a whole game from a single sentence, and this pack shows you how, step by step.

Every build from the walkthrough, a physics roulette wheel, a GTA-style open world, an exploding interactive guitar, each with the exact prompt behind it.

One prompt in, a working website out, published to a link you can share the same day.

At I/O in May, Google shipped Gemini 3.5 Flash and said 3.5 Pro was already running internally and would roll out next month.

That was May 19 and it’s almost end of July already.

On July 16 Bloomberg reported Google is months behind on Pro, specifically on coding, sourced to 10 current and former employees.

Then on July 21, Google shipped three of them.

Still no Pro.

The blog says it's "currently testing with partners," and that Google has started its most ambitious pre-training run yet, for Gemini 4.

Google's pitch for 3.6 Flash is a step up in coding and knowledge work + better token efficiency.

Artificial Analysis benchmarked it before launch, and the efficiency claim holds but the capability doesn't.

So it’s just a good product update sold as a model upgrade.

He called it the worst results he'd ever gotten on frontend and spatial reasoning, then added he hoped it wasn't running on high thinking.

That last line is the whole thing.

Google changed the default effort level from high to medium this generation.

Speed first, since that's the loudest complaint.

5 everyday prompts, of which response time was a non-issue every time except one, a movie plot question, where it sat there for a few seconds before answering.

That's the app going off to fetch a poster and sources and has nothing to do with the model straining.

The 11.5-second figure from the lab benchmarks never showed up in normal use.

Then I asked it to draw three things as SVG, and it didn't.

All three came back as generated images, which means the app handed my request to its image model rather than writing any code.

Funny but completely useless as a test of reasoning, so I'm binning it.

Now the webpage. I asked for a single-page site for a made-up Bengaluru filter coffee subscription. Hero, three pricing tiers, FAQ accordion, footer, make it look expensive.

It gave me all of it + an interactive calculator that works out your monthly grams from cups per day.

This was one-shot generation, and the people complaining were mostly deep in long coding sessions, which is a different job.

But the accusation was terrible frontend capability, and that is not what came back for me.

The Flash models are fine, but Pro is the focus now.

Google shipped something cheaper and faster that does decent work, and the reaction was closer to grief than criticism, which makes sense if you stop reading it as a review of 3.6 Flash.

People are angry at a small model because the big one was promised in June and still hasn't arrived, and Flash is the only thing standing there to be angry at.

And Gemini 4's pre-training only just started, so the question of whether Google is still a frontier lab doesn't get answered this quarter.

So until Pro lands, every Flash release reads as a stalling tactic. (deserved or not)

Until next week,
Vaibhav 🙌🏻

If you read till here, you might find this interesting

#AD 1

If you're a high earner, your portfolio could be quietly costing you thousands in avoidable taxes. On August 6, join Range's CFPs and CPAs for a live session on the tax-smart investing moves that matter most in 2026. 

You'll learn how tax-loss harvesting, asset location, direct indexing, and withdrawal sequencing can help grow your after-tax returns — and walk away with an action plan you can run immediately. Range delivers comprehensive wealth management, aligning tax planning, investment management, retirement, and estate planning into one hyper-personalized plan so nothing slips through the cracks. 

Bring your questions; our experts will answer them live. Seats are limited.

RSVP to join for free

^{This webinar is for informational purposes only and does not constitute investment advice or a recommendation to buy, hold, or sell any security. Forward-looking statements involve risks and uncertainties. Past performance is not indicative of future results. Range defines "high earners" as households with income over $300k.}

#AD 2

ChatGPT is a superpower if you know how to use it correctly.

Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.

Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.

Download the free guide.