Max

25. no wife. no kids. no property. start in tech from scratch in 2026 when everyone cries that tech is over.

[Max @maxt3chno]

Everyone is hyping up fable 5.1 and astra... but are they actually that much better?

YES they ARE.

saw a sample of fable 5.1's output compared against gpt-5.6 sol and opus 5

yeah prompts were different and idk how many shots/retries fable needed, but the detail is actually insane

timeline rumors say gemini 3.8 flash, astra, and fable 5.1 are all dropping this week

we’ll see if big tech corps can reclaim the hype or if open-weights already cooked them beyond repair lmao

[Vib3Coded @vib3coded]

This voxel comparison got way more interesting than I expected

I gave Claude Opus 5, GPT-5.6 Sol, and Gemini 3.7 Flash the same basic challenge: turn a reference image into a large, interactive voxel scene of a Japanese temple complex

And somehow, all three created completely different worlds

Claude built the tallest and most architectural version, with a huge central pagoda and a dense, polished temple district

Sol created the largest environment: 67,097 voxels, multiple temples, bridges, ponds, a waterfall, cherry blossoms, and atmospheric night lighting

Gemini took a cleaner and more minimal approach, with brighter colors, a simpler garden layout, and a strong central pagoda

Same reference. Same idea. Three completely different interpretations

I didn’t write any code. I just shared the reference, explained what I wanted, and gave a few follow-up adjustments

I put all three results into one 50-second vertical video so you can compare them side by side

Which one would you choose?👀

Prompt and reference image are in the comments

[Max @maxt3chno]

@MyDaebakCafe ahah lol yep my bad

anyway nobody should take it too seriously, it’s just a vibe test tbh. for actual benchmarks there’s other resources

[Max @maxt3chno]

Just tested Gemini 3.8 flash in antigravity on the same one-shot 3D

I'm not impressed at all

glm-5.3 flash is STILL the clear winner here, and i haven’t even tested fable 5.1 or opus yet lol

solid 6/10. yeah it’s cheap, fast af, and textures improved over 3.7, but food assets still look like low-poly generic blocks. UI works but the overall visuals are completely mid

definitely gotta run astra vs fable 5.1 on this benchmark next fr

[Max @maxt3chno]

This is actually INSANE. Can’t believe how GLM-5.3 (ex-0xAlpha) won and AGAIN beat corp models like grok, gemini

look at this API pricing per 1M tokens:

- input: $0.075
- output: $0.25
- cache: $0.015

costs literal pennies and passed the one-shot vibe check way better. solid 7/10

UI is cleaner, and while 3D food assets still look like pure shit, the lighting, logic, and zero texture clipping are actually working

[Max @maxt3chno]

but nobody said this is THE benchmark, it’s literally just a vibe check

on top of that my whole point is there’s no perfect model, every model’s good at some stuff and trash at other stuff (and read the post, i literally said glm’s a W here in that specific case lol)

don’t take it too seriously, it’s just one test outta thousands on that platform. i just like checking what ai can visually pull off rn, not really interested in like how fast it rewrites an excel file or whatever

[Max @maxt3chno]

@MyDaebakCafe what do u mean? how can a test be wrong

[Max @maxt3chno]

@Ananth7e feels like it was straight up benchmaxed fr

and i’m not even talking coding here, just asked some linux terminal questions and it took way more steps to fix the issue than it shoulda. didn’t have this problem with 3.7

[Max @maxt3chno]

@bridgemindai is anyone even surprised tho? kinda obvious google and tough coding tests just don’t mix

p.s. still think it’s the best model for the average user tho

[Max @maxt3chno]

@SPAC89 this is weird, still don’t have it even in google ai studio. maybe they only rolled it out for certain countries rn

[Max @maxt3chno]

@aapakari didn’t catch that tbh, was out on a walk and only caught like 30 mins of the whole stream

[Max @maxt3chno]

OMG. this is actually insane. fable 5.1 just completely cooked every other model

have you guys seen the outputs yet?

that same rocket launch prompt everyone’s been benchmarking: bridgeminde ran it through his setup and it is straight up PURE CINEMA fr

vfx look crazy, but the sound design is the real diff = easily the best audio generated by an ai coding-model

idk what astra can even pull off to compete with this lmao, bar is in the stratospher

[Max @maxt3chno]

WTF... 0xAlpha is actually on another level

just threw the same rocket launch one-shot prompt at it and I’m speechless

easy 9/10. and this is ALL ONE PROMPT on a FREE model (for now)

honestly can’t even find any real flaws here, maybe some slight overexposure on the launchpad model but that’s literally it

everything else is butter smooth, features work, audio goes crazy

the AI race is moving way too fast bro

[Max @maxt3chno]

@Bhavani_00007 bridgemind did a rocket launch live on stream today using fable 5.1, it was so fckng cool, looked like 10x better tbh

[Max @maxt3chno]

@JohnPay53891334 yeah honestly great idea, fable 5.1 legit made a real step up in ui

[Max @maxt3chno]

Everyone is hyping up fable 5.1 and astra... but are they actually that much better?

YES they ARE.

saw a sample of fable 5.1's output compared against gpt-5.6 sol and opus 5

yeah prompts were different and idk how many shots/retries fable needed, but the detail is actually insane

timeline rumors say gemini 3.8 flash, astra, and fable 5.1 are all dropping this week

we’ll see if big tech corps can reclaim the hype or if open-weights already cooked them beyond repair lmao

[Vib3Coded @vib3coded]

This voxel comparison got way more interesting than I expected

I gave Claude Opus 5, GPT-5.6 Sol, and Gemini 3.7 Flash the same basic challenge: turn a reference image into a large, interactive voxel scene of a Japanese temple complex

And somehow, all three created completely different worlds

Claude built the tallest and most architectural version, with a huge central pagoda and a dense, polished temple district

Sol created the largest environment: 67,097 voxels, multiple temples, bridges, ponds, a waterfall, cherry blossoms, and atmospheric night lighting

Gemini took a cleaner and more minimal approach, with brighter colors, a simpler garden layout, and a strong central pagoda

Same reference. Same idea. Three completely different interpretations

I didn’t write any code. I just shared the reference, explained what I wanted, and gave a few follow-up adjustments

I put all three results into one 50-second vertical video so you can compare them side by side

Which one would you choose?👀

Prompt and reference image are in the comments

[Max @maxt3chno]

@bridgemindai yeah benchmarks are cool and all but real capability only shows up through actual real world testing

[Max @maxt3chno]

@Haleeeemahh love how every ai ceo makes agi predictions and somehow it’s always THEIR product that’s gonna be the one lmaooo

[Max @maxt3chno]

@Ananth7e ngl i test out a bunch of chinese models regularly, they’re pretty solid too, like glm 5.3 flash for example. but noticed if i gotta tackle something actually tough, fable or opus is my go-to number 1 pick every time

[Max @maxt3chno]

@sha_sujin depends how u look at it tbh. closest to the prompt and reference is gpt fr, but ngl grok’s water shading hits different for me

[Max @maxt3chno]

This comparison got way more interesting than I expected

I gave grok 4.6 build, gpt-5.6 sol, gemini 3.7 flash, and glm-5.3 flash the exact same prompt + image ref: build an interactive arcade water sim with sharks in three.js

all 4 cooked up completely different outputs:

❌ gemini 3.7 (biggest L): output runs fine and tap works, but it completely ignored the reference image aesthetic and built it in a totally different style lol

🌊 grok 4.6: reflections look like a high-budget render and beach matches the ref. sharks look a bit scuffed tho, and ripple physics are way too aggressive

💀 glm-5.3 flash: water looks simple but realistic, sharks are decent, but tap is buggy, and it casually melted 1.5M tokens lmao

🔥 gpt-5.6 sol: by far the closest to the image ref and prompt. tap is mid, but the rest is straight fire (not a pure one-shot run ngl)

zero perfect models exist...
each has its own pros and cons, and anyone claiming ai can reliably one-shot complex stuff rn is on pure copium

[Max @maxt3chno]

@wholyv i bet fable wouldn’t even attempt this if it saw “cut” and “head” in the same sentence lmaooo

[Max @maxt3chno]

@LeMiMind looks way better than it did in full size tbh, the ratio i use to combine all 4 vids into one kinda covers up some of the flaws

[Max @maxt3chno]

@goodworse has anyone asked where the pro model is yet?google we appreciate the speed, but we need a heavy coding powerhouse too xD