GOOGLE 🔥: Gemini 4 Argon has reportedly been A/B-tested under Gemini 3.8 Flash on Antigravity for some time. It seems like a huge jump compared with the existing Flash model. World gen prompt > Gemini 4 Argon (supposedly)

World gen prompt > Current Gemini 3.8 Flash

More Gemini 4 outputs in the blog 🗞️

Still not that good at coding 🥀

Yet this was slightly better than GPT-6.1 overall, and significantly better in some areas (Which is expected, I guess)

Still isn't out yet?...

Soon 💀

@Marie_Haynes 🫣🫣

Why is there no news about open-source models anymore... 😭

Deepmind still has it (based on its hardware stack and data it was inevitable for them to release a more powerful model). we definitely can not count Google out at this point

Also under gemini 3.8 flash sometimes on the web

This is like the 8th post today hinting at the launch prep for Gemini 4. Looks like we will need to buckle up!

If a Flash-tier model is quietly beating the old Pro on world generation, the pricing ladder gets messy fast. Curious whether Argon lands first in Antigravity's agent loops or in the consumer app.

Let's see

If Argon already beats Flash in A/B, the public drop will be loud.

atp everything is being routed to supposedly stronger models that only some specific people have access to 🤔

if that's the flash one what does pro look like 💀

A huge jump from a supposedly unreleased model is a bold read

Gemini 4 Argon will be newer than Gemini 3.8 Flash, sources tell me.

a leaked model running as an A/B under an old name is how you get benchmark contamination without trying. world gen looks sharp though.

it is not coming today, so testing must be going on

silent A/B under Flash is the real launch. marketing just catches up later

Gemini 4 hiding inside a Flash costume like three raccoons in a trench coat.

好家伙 Flash 马甲里藏了个 4 🧋

the A/B test era of model launches is so funny to me — google's flagship might have been live for weeks and nobody even noticed

a 'huge jump' quietly a/b tested under the flash name is peak google launch strategy

This doesn't look that good as claude next for sure

It’s crazy how this almost does nothing for me anymore. I’ve just watched all week people making realistic m AutoCAD rendered objects in three js using opus 5.5.