Havis | Claude Code

I run Claude Code all day. posting the setups that actually ship.

[Havis | Claude Code @HavisT83099]

https://t.co/IPql4GGR6x

[Havis | Claude Code @HavisT83099]

@buzzhive_ai the swinging doors part is such a fun test 😂

[Havis | Claude Code @HavisT83099]

@buzzhive_ai the monitor was the part i didn't expect at all, it just sinks right into the desk like it was always water

[Havis | Claude Code @HavisT83099]

HOLY SH*T. I GAVE CLAUDE AND GEMINI ONE PICTURE OF MY DESK AND TOLD THEM TO TURN IT INTO WATER

Claude Opus 5.5 worked 15 minutes. $2.44.
Gemini 4 Argon worked 2.5 minutes. $0.13. 19x cheaper.

Which one would you put on your desk?

[Havis | Claude Code @HavisT83099]

https://t.co/IPql4GGR6x

[Havis | Claude Code @HavisT83099]

@buzzhive_ai same, that waterfall is what sold the whole thing

[Havis | Claude Code @HavisT83099]

@Meddes0119 appreciate it, the crossed out name still gets me

[Havis | Claude Code @HavisT83099]

Claude Opus 5.5 and GPT-6 Astra made an ad telling you to switch to their rival.

Claude crossed out its own name and said "switch today".

GPT worked twice as long and asked "what if you switched?". Then added "not a live product capture".

[Havis | Claude Code @HavisT83099]

https://t.co/IPql4GGR6x

[Havis | Claude Code @HavisT83099]

@buzzhive_ai 4.8 holds up better than people say tbh, 5.5 just piles on way more little details

[Havis | Claude Code @HavisT83099]

IS OPUS 4.8 REALLY THAT MUCH F*CKING WORSE?

Everyone says Opus got nerfed. So I gave 3 generations the same picture and the same prompt: "bring this cozy room to life."

Top to bottom: 5.5, 5, 4.8.

[Havis | Claude Code @HavisT83099]

https://t.co/IPql4GGR6x

[Havis | Claude Code @HavisT83099]

@buzzhive_ai penguins that sulk then forgive are too cute 😂

[Havis | Claude Code @HavisT83099]

WHAT THE F*CK. CLAUDE TURNED 4 SCREENSHOTS OF A GAME THAT DOESN'T EXIST INTO REAL GAMEPLAY

I generated the screenshots. Same prompt for both.

/ Claude: sunlight, jellyfish, a crystal cave, 5/5 pearls, an anglerfish at 2,890 m.
/ Grok: 20 minutes of thinking for some grey rocks and a black screen.

[Havis | Claude Code @HavisT83099]

https://t.co/IPql4GGR6x

[Havis | Claude Code @HavisT83099]

@HodlMiro jumping off the golden gate and just flying back up is insane 😂

[Havis | Claude Code @HavisT83099]

@pantherbllack "it can't show up for you" is the best line here 😂

[Havis | Claude Code @HavisT83099]

@InTheAssembly worth reading past the headline, most of that loss is a non-cash charge on the notes that grows with valuation

[Havis | Claude Code @HavisT83099]

@testingcatalog a proactive agent built specifically for work is the part that gets me, most personal agents still wait for you to ask, curious how it decides when to step in

[Havis | Claude Code @HavisT83099]

Claude Opus 5.5 and GPT-6 Astra made a launch video for each other's next model.

GPT called it Claude Next.

Claude worked for 3 minutes and announced GPT-6. The model sitting right next to it.

[Havis | Claude Code @HavisT83099]

HOLY SH*T. I ASKED CLAUDE AND GEMINI TO BUILD A ROCKET CAR JUMP. SAME PROMPT

Car boosts on rear thrusters, hits one ramp, flies over 4 cars and a truck. No landing ramp. Jets under the car fire down and set it on the ground soft

Zero edits. Code straight into the browser:

/ Claude Opus 5.5: full stunt in 12 seconds, landing jets and dust, done
/ Gemini 4 Argon: the sickest boost of the two, then hangs in slow-mo so long the landing barely makes the cut

Google's newest flagship vs Anthropic's. Which one would you put in the driver's seat?

[Havis | Claude Code @HavisT83099]

HOLY SH*T. 3 AI MODELS, SAME PROMPT, 3 WORKING WATCHES. ONE COSTS 6X MORE

The prompt: a luxury skeleton watch in 3D. Gold case, ticking hands, gears turning behind the dial

Zero edits. Code straight into the browser:

/ Claude Opus 5.5 (~$3 per try): real-looking leather strap with clean stitching, gold that actually shines
/ GPT-6 Astra (~$0.50 per try): rough strap, sloppy stitching, flat one-tone gold
/ Kimi K3 (~$0.60 per try): cartoonish strap, weak stitching, tried to make the gold sparkle and didn't quite get there

The gears spin fine on all 3

A year ago none of them could do this. Now the only question is what the extra $2.50 buys you

[Havis | Claude Code @HavisT83099]

It's getting f*cking scary

A solo dev in Shenzhen just wired up almost 300 Claude Code agents running at once, plus a second layer whose only job is to catch them lying

And the dashboard is insane:
/ ~280 agents writing and fact-checking at the same time
/ every single claim checked against its source in under a second
/ 120 company profiles, 3 passes, zero errors left
/ all of it dropping straight into his Obsidian vault

The trick is stupid simple. A big model does the work, a cheap one only decides if each claim is actually true. The loop doesn't stop until nothing fails

A research team would bill you weeks for this. He did it overnight. From his bedroom

The whole setup, step by step, is in the guide below

[draven @dravenip]

https://t.co/ekWOSQOOZH