orvian

just a curious guy exploring what AI can do

[orvian @heyorvian]

@Rivonn thank you Rivon

[orvian @heyorvian]

@kawaidgaf thank you fren

[orvian @heyorvian]

@BuzzMalx yeah yeah, i need to do another test

[orvian @heyorvian]

Anthropic’s cheapest model rn vs OpenAI’s cheapest model

Claude Haiku 5.5 vs GPT-6 Luna

same prompt, different output

who won?

[ClaudeDevs @ClaudeDevs]

Haiku 5.5 is now available in the Claude Platform and Claude Code. On average, it costs around 75% less to run than Haiku 4.5.

It pairs well with Opus 5.5 or Sonnet 5.5 as a subagent. Use it for high-volume, cost-sensitive tasks like summaries, compactions, or database queries.

[orvian @heyorvian]

@kaddisdeployed hahaha

[orvian @heyorvian]

read more: https://t.co/RuE5Oz9AKB

[orvian @heyorvian]

a hacker allegedly used AI agents to attack South Korean banks

then made one ridiculous mistake 😭

they left their Claude Code session histories exposed on a public server.

CrowdStrike researchers found the logs alongside agent memory files and configurations for ARTEX, an open-source AI penetration-testing tool.

the attacker reportedly used DeepSeek V4.1 Flash, GLM-5.3 and Grok 4.6 across the operation.

and somehow, it gets worse.

in one Claude Code session, the attacker asked AI to write a security researcher résumé, providing personal details that could help identify them.

researchers also found conversations about where stolen Korean data could be sold.

the attacker’s identity hasn’t been confirmed.

imagine using AI to help hack banks, only to potentially expose yourself through your own AI chat history

[orvian @heyorvian]

@nex_ify great, thank you Nexx

[orvian @heyorvian]

@nex_ify i honestly think people are getting more stupid already 😭😂

[orvian @heyorvian]

@sashikantsingh_ claude is cooking gpt so bad

[orvian @heyorvian]

@painn_x @envato testing it, i love this

[painn @painn_x]

if you're still generating one image per idea, you're doing it the slow way

i used to write a prompt, hate the result, rewrite it, hate that too.

same loop for like 20 mins and a bunch of credits gone

Burst mode by @envato does up to 6 directions for 1 credit.

you just look at them side by side and go with whatever's actually good

and if one is close but not quite there you can ask for more like it instead of starting over

https://t.co/dPyMOhqIVM

[orvian @heyorvian]

@MAXdeg0 damn, gotta try this

[orvian @heyorvian]

@MAXdeg0 yeah, gpt won this round

i’ll try to do another test

[orvian @heyorvian]

@MrTacticalX opus won😭

[orvian @heyorvian]

@Mr_Salio let me try it asap

[orvian @heyorvian]

@ravikiran_dev7 Elon also left openAI

[orvian @heyorvian]

@notjazii how is this possible 😭

[orvian @heyorvian]

@painn_x joined the waitlist

[orvian @heyorvian]

@kaddisdeployed beating openai so much rn 😭

[orvian @heyorvian]

Anthropic just launched its cheapest Claude model yet

Claude Haiku 5.5 costs around 75% less per task than Haiku 4.5

and up to 90% less per token for workloads under 100K tokens, according to Anthropic.

It’s built for fast, high-volume workloads like:

→ Coding sub-agents
→ Browser agents
→ Customer-facing agents
→ Real-time apps
→ Tasks where larger Claude models get expensive

Haiku 5.5 also introduces adjustable effort, letting developers control how much reasoning each task uses.

Available now through the Claude app, API, AWS, Google Cloud, and Azure.

This could make multi-agent systems far cheaper to run at scale

[ClaudeDevs @ClaudeDevs]

Haiku 5.5 is now available in the Claude Platform and Claude Code. On average, it costs around 75% less to run than Haiku 4.5.

It pairs well with Opus 5.5 or Sonnet 5.5 as a subagent. Use it for high-volume, cost-sensitive tasks like summaries, compactions, or database queries.