this is f**king insane
xiaomi just cooked openai on a coding task
i gave them the same prompt, mimo took longer but the 3d it shipped is crazy
watch the side by side
codex users, you need to see this... 🚨
xiaomi's model just out-built openai on the same prompt
i gave both one task: code an animated 3d solar system in a single html file
> gpt-6.1-terra-low: done in ~30 secs
> mimo-v2.6-pro: ~55 secs, but built way more
ran both pages. here's what mimo shipped:
- real 3d you can drag to rotate
- saturn's ring + twinkling starfield
- all 8 planets labelled + a speed slider
terra kept it clean: tilted 2.5d orbits, labelled planets, speed slider
speed → terra
features → mimo
which one are you picking?
codex users, you need to see this... 🚨
xiaomi's model just out-built openai on the same prompt
i gave both one task: code an animated 3d solar system in a single html file
> gpt-6.1-terra-low: done in ~30 secs
> mimo-v2.6-pro: ~55 secs, but built way more
ran both pages. here's what mimo shipped:
- real 3d you can drag to rotate
- saturn's ring + twinkling starfield
- all 8 planets labelled + a speed slider
terra kept it clean: tilted 2.5d orbits, labelled planets, speed slider
speed → terra
features → mimo
which one are you picking?
nyro retweeted
another powerful DeepSeek v4.1 jailbreaking tool?
It’s called Pojia Next, and it’s completely free & open-source
Instead of being stuck with the default instructions inside DeepSeek’s coding agent, it lets you replace them with your own
It can:
• Modify built-in prompts
• Apply changes with one click
• Automatically restart the client
• Back up original files before making changes
• Restore everything if something goes wrong
It also supports other AI clients
The repo has already crossed 900 GitHub stars
save this and start using now!
DeepSeek just open-sourced another crazy AI coding agent 😭
this one can handle multiple GitHub tasks at once
it’s called dsh-superagent, and instead of babysitting your coding agent in the terminal, you can assign tasks through GitHub and let it work
it can:
• run multiple coding sessions in parallel
• work on issues and generate pull requests
• keep tasks isolated from each other
• let you review progress through a web dashboard
it’s completely open source and already crossed 2K GitHub stars, bookmark and start using it now!
opencode and claude users, grab this now before it's gone...
a mystery model just went FREE for 7 days 😳
nobody knows who made it
> 1M context
> reads text, images, audio AND video
> $0 on openrouter
> providers say no training on your data
here's how to set it up in 2 mins:
- open opencode
- type /models
- pick crouton alpha, done
already on openrouter? the model id is stealth/crouton-alpha
note: opencode says it's region-locked in some places, and some people are hitting daily caps
give this to your agent now. use it while it lasts
openrouter link here:
it's also free in command code and perch ai 👀
teapawt.org/OpenRouter/status/2117…
nyro retweeted
STOP JOB HUNTING MANUALLY!!!
someone just open-sourced an AI job hunting assistant
it’s called AI Job Search, and it runs with Claude Code
the repo already has over 67k stars and 21k forks
it can help you:
• find job openings that match your skills
• rank jobs based on your experience
• tailor your CV for each role
• write personalized cover letters
• prepare for interviews
• track applications and follow-ups
the dev used it to send 69 applications, land 20 first interviews and eventually get hired as an AI engineer
it’s free and open-source on GitHub
holy sh*t this is f**king dangerous
a food delivery app just dropped a giant coding model
LongCat-2.5 by Meituan has a 1M token context window
it outputs up to 256K tokens in one go
new accounts reportedly get 10M free tokens
and it plugs straight into Claude Code
[here is how you set it up]
- sign up on the LongCat API platform and create an API key
- set ANTHROPIC_BASE_URL to api.longcat.chat/anthropic
- set ANTHROPIC_AUTH_TOKEN to your key, run claude. done that's it!
save this no matter what. this will be the best thing you do this week
Introducing LongCat-2.5, our latest natively multimodal agentic model with 1.6T total parameters and ~48B active.
Built for long-horizon agentic tasks, it brings stronger visual understanding and agentic coding to work ranging from complex coding projects to visual coding such as creative design and 3D tasks.
Learn more about LongCat-2.5: longcat.ai/blog/longcat-2.5/
kimi-k3 vs gpt-6.1-terra-low... this was not close 👀
same prompt: neon snake game, one html file, score counter, speed ramp, restart button
terra finished in ~25 seconds
kimi-k3 thought for 2m 37s... then cooked
here's what each one shipped:
- terra → neon grid, score, restart, on-screen arrows
- kimi-k3 → flickering neon title, glowing borders, score + speed readout, keyboard + swipe controls
speed goes to terra
features go to kimi-k3
which one are you picking? 👇
Sam Altman when he hears your phone has 8gb RAM
Sam Altman said your PC's RAM is a waste
And as you expect, Elon Musk had a rebuttal
Sam Altman:
"But from a global allocation perspective, actually, it's quite terrible to have everybody have a bunch of the world's limited supply of DRAM sitting in their own computer. And, being used 5% of the time."
"You want maximum utilization in the cloud if you're trying to deal with this very supply constrained environment."
Elon Musk:
"The people deserve to have their RAM too!"
Sam Altman said your PC's RAM is a waste
And as you expect, Elon Musk had a rebuttal
Sam Altman:
"But from a global allocation perspective, actually, it's quite terrible to have everybody have a bunch of the world's limited supply of DRAM sitting in their own computer. And, being used 5% of the time."
"You want maximum utilization in the cloud if you're trying to deal with this very supply constrained environment."
Elon Musk:
"The people deserve to have their RAM too!"
this is f**king insane
freellmapi pools free tiers from 34 providers into one local endpoint
45k stars, automatic failover when one hits its limit
keys stay encrypted on your machine, it claims ~7.4b free tokens a month across all of them
here is how you set it up:
- npx freellmapi
- add free keys from groq, openrouter, nvidia, gemini etc in the dashboard
- npx freellmapi setup-claude
save and bookmark this no matter what
github.com/tashfeenahmed/fre… here is the GitHub repo
this is f**king insane
Vercel Labs just open-sourced agent-desktop, it lets any coding agent (Claude Code, Codex, OpenCode, whatever) actually control real desktop apps on macOS, Windows and Linux
it reads the OS accessibility tree instead of guessing from screenshots so actions are stable and retry-safe
here is how you set it up:
- npm install -g agent-desktop
- grant accessibility permissions (agent-desktop permissions –guide)
- drop the skill into your agent or just tell it to use agent-desktop
save and bookmark this no matter what
github.com/vercel-labs/agent… here is the GitHub repo
this is f**king insane
Vercel Labs just open-sourced agent-desktop, it lets any coding agent (Claude Code, Codex, OpenCode, whatever) actually control real desktop apps on macOS, Windows and Linux
it reads the OS accessibility tree instead of guessing from screenshots so actions are stable and retry-safe
here is how you set it up:
- npm install -g agent-desktop
- grant accessibility permissions (agent-desktop permissions –guide)
- drop the skill into your agent or just tell it to use agent-desktop
save and bookmark this no matter what
github.com/vercel-labs/agent… here is the GitHub repo
this is f**king insane
Mistral just open-sourced the weights for Large 4 (Le Chonk)
1 trillion parameters, 52B active, and it hits 41% on Terminal-Bench 4.0 for coding agents, it ties the best open-weight models while using 47% fewer tokens per task
here is how you set it up:
- download the weights from Hugging Face
- run it with vLLM or SGLang (or use the API at $0.70/$2.10 per million tokens)
- point your coding agent at it
save and bookmark this no matter what
here is the GitHub / Hugging Face repo huggingface.co/mistralai/Mis…
nyro retweeted
WTF, someone just dropped a FREE jailbreak prompt for DeepSeek V4.1 😳
and apparently, it works with GPT models too
-> copy, paste and run
-> no installation needed
-> publicly available on GitHub
repo link in cs
use it now while it lasts
okay, this is F**KING GOLD for game developers
someone just open-sourced an AI agent that can play video games and find bugs on its own
it’s called Aura Autopilot, built by Ramen
you can give it a task like defeating a boss, exploring a level or testing a feature, and it’ll play the game itself and generate a report of what went wrong
it supports Unreal Engine, Unity and Godot, and works with Claude Code and Codex
and it’s completely open source
this is f**king insane
Mistral just open-sourced the weights for Large 4 (Le Chonk)
1 trillion parameters, 52B active, and it hits 41% on Terminal-Bench 4.0 for coding agents, it ties the best open-weight models while using 47% fewer tokens per task
here is how you set it up:
- download the weights from Hugging Face
- run it with vLLM or SGLang (or use the API at $0.70/$2.10 per million tokens)
- point your coding agent at it
save and bookmark this no matter what
here is the GitHub / Hugging Face repo huggingface.co/mistralai/Mis…
this is f**king insane
I cancelled my $200/mo Claude for this, a free github repo with 65k stars just dropped the entire proxy that lets you run Claude Code, Codex and OpenCode against 61 free providers
you get 1.3B+ free tokens a month and it auto-falls back if one dies
here is how you set it up:
- curl the install script from the repo
- point it at your free keys (NVIDIA NIM, OpenRouter, local, etc)
- run fcc-claude instead of claude
save this and give it to your agent now!
github.com/Alishahryar1/free… here is the GitHub repo
this is f**king insane
I cancelled my $200/mo Claude for this, a free github repo with 65k stars just dropped the entire proxy that lets you run Claude Code, Codex and OpenCode against 61 free providers
you get 1.3B+ free tokens a month and it auto-falls back if one dies
here is how you set it up:
- curl the install script from the repo
- point it at your free keys (NVIDIA NIM, OpenRouter, local, etc)
- run fcc-claude instead of claude
save this and give it to your agent now!
github.com/Alishahryar1/free… here is the GitHub repo
this is f**king useful
apple is bringing AI agents directly into pages, keynote and numbers
→ write and edit documents
→ create and improve presentations
→ work with data, formulas and charts
→ connect external AI agents through MCP
your AI can now do more than just chat with you
it can actually help get the work done.
APPLE IS BRINGING AI AGENTS INTO PAGES, KEYNOTE AND NUMBERS
you can now get AI help inside your documents, presentations and spreadsheets
- Pages: write and edit your content
- Keynote: create and improve slides
- Numbers: work with data, formulas and charts
- MCP lets external AI agents read and edit files inside these apps
how to try it:
update your Apple apps when the new features become available
open a document, slide deck or spreadsheet
use the built-in AI assistant to work on it
on Mac, MCP support lets compatible external agents connect too
the assistant is in beta, with access requirements and language limits
this could make everyday work a lot easier
you trying this?
there is f**king free coding credits you're sleeping on 😳
openrouter has a whole free models list
Nemotron 3 Ultra hits $0/M in + out, 1M context, while you're still burning a $20/mo claude plan for the same "write me a function" prompts
→ search free on openrouter
→ open the one that says (free) / Price: Free
→ chat it → reply shows $0
under 3 mins. screen recording is the whole path.
go milk it before they throttle the free tier 🚨
openrouter.ai/models?q=free
openrouter.ai/nvidia/nemotro…
this is f**king insane
openai just dropped gpt-6.1 terra
it looks at your screen, clicks through apps, and does the boring work for you
→ 50% cheaper than the last terra
→ matches gpt-6 astra at 1/10 the price
→ 3x faster
→ live in chatgpt, codex and the api right now
the cheap model just learned to use a computer
use this before everyone else does
