Video wird geladen...

Video konnte nicht geladen werden

Beim Laden dieses Videos ist ein Problem aufgetreten. Dies könnte an einem vorübergehenden Netzwerkproblem liegen oder das Video ist möglicherweise nicht verfügbar.

This is a 9B model running a full agentic loop with Wi-Fi off. Ornith 1.5. It read three sheets, cross-referenced the monthly rollups, and wrote a revenue analysis on its own. Nothing left the Mac.

Osaurus

10,803 subscribers

104,455 Aufrufe • vor 2 Monaten •via X (Twitter)

Wissenschaft & Technologie

Anya Rossi• Live Now

Private livecam show

31 Kommentare

Profilbild von Osaurus

Osaurusvor 2 Monaten

Ornith 1.5 9B, MXFP8 quant on MLX. Recorded with Wi-Fi off, start to finish. Ornith reports the 9B matches or beats Gemma 4 31B and Qwen 3.6 35B. Watching it plan, read files, and self-correct, that tracks.

Profilbild von Osaurus

Osaurusvor 2 Monaten

Free and open source. MIT licensed. Native macOS, not Electron.

Profilbild von Ornith

Ornithvor 2 Monaten

🧡thanks for your great work!

Profilbild von Osaurus

Osaurusvor 2 Monaten

Thank you for the great model!

Profilbild von Aleksandar Janca

Aleksandar Jancavor 2 Monaten

how fast is it?

Profilbild von Osaurus

Osaurusvor 2 Monaten

It’s decently fast, depends on your Mac

Profilbild von 师梦豪

师梦豪vor 2 Monaten

@Jungian_coach 🫶

Profilbild von Osaurus

Osaurusvor 2 Monaten

@Jungian_coach ❤️

Profilbild von Berlioz

Berliozvor 2 Monaten

I never seem to get the same results you guys do when I try osaurus on my own

Profilbild von Osaurus

Osaurusvor 2 Monaten

Which model did you try?

Profilbild von Berlioz

Berliozvor 2 Monaten

Tried many models including the previous Ornith ones. Even GPT 5.6 Sol frequently fails basic tool calls

Profilbild von Osaurus

Osaurusvor 2 Monaten

With the latest version?

Profilbild von Berlioz

Berliozvor 2 Monaten

There’s been some updates since the last attempts, will give it another shot

Profilbild von Martin Ronfort

Martin Ronfortvor 2 Monaten

That local execution is what makes the 9B so compelling for real work. We actually broke down what this model can do here:

Profilbild von George Mastorakos, MD

George Mastorakos, MDvor 2 Monaten

wow! what hardware?

Profilbild von Osaurus

Osaurusvor 2 Monaten

Mine is MacBook Pro M4 48GB Ram

Profilbild von RiftStack AI

RiftStack AIvor 2 Monaten

a 9B holding an agent loop together is a big deal. did you have to cut the tool set down to keep it reliable, or constrain it some other way?

Profilbild von Osaurus

Osaurusvor 2 Monaten

It’s a delicate balance but keeping contexts minimal is the key to success

Profilbild von deX

deXvor 2 Monaten

wifi off is the detail that changes everything

Profilbild von Fanfulla

Fanfullavor 2 Monaten

9b doing arithmetic across three sheets, i'd check the totals

Profilbild von Rick | DMKTZ | FITzOn

Rick | DMKTZ | FITzOnvor 2 Monaten

offline 9b model running full agent loops without leaving the mac. privacy is nice but i doubt it can beat a cloud api at actual reasoning

Profilbild von Cuth

Cuthvor 2 Monaten

On a macbook pro m5 24 gb?

Profilbild von scoraclescotty

scoraclescottyvor 2 Monaten

Really good work, guys. This kind of experience is the next frontier.

Profilbild von Osaurus

Osaurusvor 2 Monaten

Thank you! We still have a long way to go.

Profilbild von ZK

ZKvor 2 Monaten

@HENKOWISH

Profilbild von 刘朝 Zhao Liu

刘朝 Zhao Liuvor 2 Monaten

A 9B model completing a spreadsheet analysis with Wi-Fi disabled is a strong privacy demo. The next proof point is reproducibility: task recipe, failure cases, and the boundary between local tools and model state.

Profilbild von AI Mastery Guide

AI Mastery Guidevor 2 Monaten

Wifi off and it still worked, wild

Profilbild von QuietLayer Studio

QuietLayer Studiovor 2 Monaten

why are you posting time lapsed video? Why not show the real speed?

Profilbild von Martino Wong

Martino Wongvor 2 Monaten

What model(s) would you suggest for osaurus with around 24/32GB memory?

Profilbild von Osaurus

Osaurusvor 2 Monaten

Try this model, Ornith 9B

Profilbild von Sebastian Buzdugan

Sebastian Buzduganvor 2 Monaten

offline is useful, but spreadsheet agents need cell-level citations before production use

Ähnliche Videos

Congrats to the Ornith team. Ornith-1.5-122B is a monster. It’s in Osaurus on day one. Running on a Mac, no cloud, no subscription. One prompt. It built a flying car game.

Congrats to the Ornith team. Ornith-1.5-122B is a monster. It’s in Osaurus on day one. Running on a Mac, no cloud, no subscription. One prompt. It built a flying car game.

Osaurus

14,411 Aufrufe • vor 10 Tagen

A 9B model just checked my Amazon shopping cart. Opened a real browser. Signed in. Read the page. Reported back. Running locally on my Mac.

A 9B model just checked my Amazon shopping cart. Opened a real browser. Signed in. Read the page. Reported back. Running locally on my Mac.

Osaurus

30,391 Aufrufe • vor 3 Monaten

This is what open weights are for. Ornith 1.0 9B running on my Mac, answering my team in Slack.

This is what open weights are for. Ornith 1.0 9B running on my Mac, answering my team in Slack.

Osaurus

16,443 Aufrufe • vor 3 Monaten

SOMEONE LEFT A MAC MINI ALONE WITH FULL COMPUTER ACCESS AND IT FOUND A $731K POLYMARKET WALLET ON ITS OWN.

SOMEONE LEFT A MAC MINI ALONE WITH FULL COMPUTER ACCESS AND IT FOUND A $731K POLYMARKET WALLET ON ITS OWN.

0xMarioNawfal

87,750 Aufrufe • vor 7 Monaten

Ornith-1.0-9B for agentic coding? It broke on tool calling mid-task and got stuck in a loop with no way to continue. Only real upside: it's less bad than Gemma-4-12B... but that's not saying much. TL;DR: Skip it.

Ornith-1.0-9B for agentic coding? It broke on tool calling mid-task and got stuck in a loop with no way to continue. Only real upside: it's less bad than Gemma-4-12B... but that's not saying much. TL;DR: Skip it.

Mia

25,969 Aufrufe • vor 4 Monaten

Wait Xiaomi has also released the best 9B model in the world?! This model is based on Qwen-9B... And you can easily run it even on 8GB of memory! - MIT license - Multimodal (image + text) - Solid for coding/agentic tasks Perfect to use locally and offline... even on a laptop with unified memory!

Wait Xiaomi has also released the best 9B model in the world?! This model is based on Qwen-9B... And you can easily run it even on 8GB of memory! - MIT license - Multimodal (image + text) - Solid for coding/agentic tasks Perfect to use locally and offline... even on a laptop with unified memory!

Paul Couvert

85,496 Aufrufe • vor 1 Monat

Gautam Bose and I left Google robotics three years ago with a simple bet: agents would be able to use the computer better than we do. and it's happening: today, we are launching Ace: an agentic teammate with its own mac mini, email, iPhone and a seat on your team. It uses the tools like you already do. no new apps to learn, no APIs to build, nothing to migrate.

Gautam Bose and I left Google robotics three years ago with a simple bet: agents would be able to use the computer better than we do. and it's happening: today, we are launching Ace: an agentic teammate with its own mac mini, email, iPhone and a seat on your team. It uses the tools like you already do. no new apps to learn, no APIs to build, nothing to migrate.

lucas ochoa

448,982 Aufrufe • vor 2 Monaten

We shrank a frontier model to 102GB. DeepSeek V4 Flash 0731 now runs on a 128GB Mac. One prompt, it wrote Minesweeper, checked its own code with Node, shipped a working game. Free on Osaurus.

We shrank a frontier model to 102GB. DeepSeek V4 Flash 0731 now runs on a 128GB Mac. One prompt, it wrote Minesweeper, checked its own code with Node, shipped a working game. Free on Osaurus.

Osaurus

143,469 Aufrufe • vor 2 Monaten

72 hours in and GPT-5.6 Sol is still running without a single follow-up prompt. The task was to build the most comprehensive map possible of the physical infrastructure powering AI, submarine cables, data centers, power plants, substations and electricity grids. The only thing given was a simple loop: → Find the biggest gap → Research it → Build it → Test it → Publish it → Repeat Most people expected this to stall after a few hours. It didn't. The Mac restarted mid-run and Codex picked up exactly where it left off without losing anything. What this actually proves is that the prompt matters less than the goal. The smarter the model, the more room you need to give it to make its own decisions. Constrain it with steps and you get a worse version of what it could do on its own.

72 hours in and GPT-5.6 Sol is still running without a single follow-up prompt. The task was to build the most comprehensive map possible of the physical infrastructure powering AI, submarine cables, data centers, power plants, substations and electricity grids. The only thing given was a simple loop: → Find the biggest gap → Research it → Build it → Test it → Publish it → Repeat Most people expected this to stall after a few hours. It didn't. The Mac restarted mid-run and Codex picked up exactly where it left off without losing anything. What this actually proves is that the prompt matters less than the goal. The smarter the model, the more room you need to give it to make its own decisions. Constrain it with steps and you get a worse version of what it could do on its own.

0xMarioNawfal

61,751 Aufrufe • vor 3 Monaten

Andrew Ng, on the biggest shift in AI right now: “The most important trend is agentic AI workflows.” In one slide, he showed the part most people still miss: GPT-3.5 scored 48% on a coding benchmark, GPT-4 reached 67%, but GPT-3.5 inside an agentic workflow jumped to around 95%. That is not a prompt trick. It is the model planning, using tools, checking its own work, fixing mistakes and running the loop again. Everyone is still arguing about which model is smarter, while the real edge is moving to the system wrapped around the model. Watch the clip, then build the workflow around your own process below.

Andrew Ng, on the biggest shift in AI right now: “The most important trend is agentic AI workflows.” In one slide, he showed the part most people still miss: GPT-3.5 scored 48% on a coding benchmark, GPT-4 reached 67%, but GPT-3.5 inside an agentic workflow jumped to around 95%. That is not a prompt trick. It is the model planning, using tools, checking its own work, fixing mistakes and running the loop again. Everyone is still arguing about which model is smarter, while the real edge is moving to the system wrapped around the model. Watch the clip, then build the workflow around your own process below.

Gipp 🦅

17,267 Aufrufe • vor 4 Monaten

i gave a 9B model an agentic OS and told it to build Octopus Invaders from scratch. Qwen 3.5 9B running through Hermes Agent with full tool access on a single 3060 12GB. browser, file system, code execution, terminal. it reads the prompt, plans the architecture, writes every file, and serves the game. one prompt, zero steering, 11 minutes. i've been going back and forth with this model for days now. text, code, reasoning. it handles small chats, batching, bash scripting. at 9B it's fast enough to feel interactive and genuinely useful for local workflows where you'd normally reach for an API. but this is the real test. can it architect and build a full game autonomously without a single correction? it's coding right now. results coming next.

i gave a 9B model an agentic OS and told it to build Octopus Invaders from scratch. Qwen 3.5 9B running through Hermes Agent with full tool access on a single 3060 12GB. browser, file system, code execution, terminal. it reads the prompt, plans the architecture, writes every file, and serves the game. one prompt, zero steering, 11 minutes. i've been going back and forth with this model for days now. text, code, reasoning. it handles small chats, batching, bash scripting. at 9B it's fast enough to feel interactive and genuinely useful for local workflows where you'd normally reach for an API. but this is the real test. can it architect and build a full game autonomously without a single correction? it's coding right now. results coming next.

Sudo su

65,570 Aufrufe • vor 7 Monaten

Run Orpheus — the state-of-the-art TTS — on your local PC! Orpheus-FastAPI is a high-performance TTS app with OpenAI-compatible API, running efficiently with LM Studio backend. And Death wrote this 1-click launcher for us! Cross platform (Mac, Linux, Windows)

Run Orpheus — the state-of-the-art TTS — on your local PC! Orpheus-FastAPI is a high-performance TTS app with OpenAI-compatible API, running efficiently with LM Studio backend. And Death wrote this 1-click launcher for us! Cross platform (Mac, Linux, Windows)

cocktail peanut

30,930 Aufrufe • vor 1 Jahr

underrated gems in Kimi-K3 release: > an early K3 wrote the majority of the kernels in the late development stages > it built a triton-class compiler from scratch, MiniTriton, that delivers performance on par with or better than Triton and torch.compile > then it designed a chip, by a model, for a model, in one 48-hour autonomous run the model is rewriting and optimizing every layer of the stack it runs on: kernels for its own training. A compiler for its own kernels. silicon for its own weights. we're watching software build its own hardware

underrated gems in Kimi-K3 release: > an early K3 wrote the majority of the kernels in the late development stages > it built a triton-class compiler from scratch, MiniTriton, that delivers performance on par with or better than Triton and torch.compile > then it designed a chip, by a model, for a model, in one 48-hour autonomous run the model is rewriting and optimizing every layer of the stack it runs on: kernels for its own training. A compiler for its own kernels. silicon for its own weights. we're watching software build its own hardware

Alex Ker 🔭

311,780 Aufrufe • vor 3 Monaten

"Tell me about the sales actuals." It opened the folder, read the spreadsheet, and broke down Q1 revenue by region. A 2.6B model. 2.3 GB on disk. Running in Osaurus, day one.

"Tell me about the sales actuals." It opened the folder, read the spreadsheet, and broke down Q1 revenue by region. A 2.6B model. 2.3 GB on disk. Running in Osaurus, day one.

Osaurus

34,708 Aufrufe • vor 2 Monaten

Anthropic engineer: "The fastest way to make an agent better isn't a smarter model, it's the right loop and graph around it." In 60 minutes she breaks down how Anthropic engineers use Claude Code, build agents that catch their own mistakes and improve with every run. This beats any $500 agentic engineering course you can find. Watch it, then read the full guide on loops and graphs below.

Anthropic engineer: "The fastest way to make an agent better isn't a smarter model, it's the right loop and graph around it." In 60 minutes she breaks down how Anthropic engineers use Claude Code, build agents that catch their own mistakes and improve with every run. This beats any $500 agentic engineering course you can find. Watch it, then read the full guide on loops and graphs below.

Anatoli Kopadze

136,422 Aufrufe • vor 2 Monaten

Can an iPhone hold its own against a Mac? 💻📱 Ollama is running DeepSeek R1 Distill Llama 8B on a 16GB M2 MacBook Air, while the iPhone 15 Pro (8GB A17 Pro) runs the same model. Same model. Same sampling temperature. Same system prompt. While the iPhone is a bit slower, the 3-bit OmniQuant model reasons just as well as the 4-bit RTN model on the Mac. 🤯

Can an iPhone hold its own against a Mac? 💻📱 Ollama is running DeepSeek R1 Distill Llama 8B on a 16GB M2 MacBook Air, while the iPhone 15 Pro (8GB A17 Pro) runs the same model. Same model. Same sampling temperature. Same system prompt. While the iPhone is a bit slower, the 3-bit OmniQuant model reasons just as well as the 4-bit RTN model on the Mac. 🤯

Private LLM

26,544 Aufrufe • vor 1 Jahr

watch and let this sit for a second anon, a 35B AI model, building a complete multi-file game by itself on hermes agent. no cloud or api. not one line of code written by a human. running on a DGX spark in 2026. because not long ago this was science fiction. this is Ornith, a new open agentic coding model, running near lossless on a single DGX Spark, a 128GB box that sits on your desk. i hand it one spec and step back. it reasons about the architecture, creates the project structure, writes every module one by one, and here's the part that got me, it checks its own work as it goes. makes a call, reads the result, decides if it's right, corrects itself before moving on. no steering from me. this is what people keep underestimating about local AI. they hear "local" and think weaker, slower, a toy. watch the clip. a 35B at near full precision, planning and self correcting on a single desktop. the open models got genuinely good at agentic work, not benchmark good, actually builds the thing good. sped 5x so you can watch the whole build start to finish. this is where local AI actually is now. not coming soon. here.

watch and let this sit for a second anon, a 35B AI model, building a complete multi-file game by itself on hermes agent. no cloud or api. not one line of code written by a human. running on a DGX spark in 2026. because not long ago this was science fiction. this is Ornith, a new open agentic coding model, running near lossless on a single DGX Spark, a 128GB box that sits on your desk. i hand it one spec and step back. it reasons about the architecture, creates the project structure, writes every module one by one, and here's the part that got me, it checks its own work as it goes. makes a call, reads the result, decides if it's right, corrects itself before moving on. no steering from me. this is what people keep underestimating about local AI. they hear "local" and think weaker, slower, a toy. watch the clip. a 35B at near full precision, planning and self correcting on a single desktop. the open models got genuinely good at agentic work, not benchmark good, actually builds the thing good. sped 5x so you can watch the whole build start to finish. this is where local AI actually is now. not coming soon. here.

Sudo su

30,299 Aufrufe • vor 4 Monaten

First look at Grok Build by xAI, this time as a CLI tool running directly in the terminal. This is clearly positioned as a competitor to Claude Code, bringing agentic coding workflows into the command line with xAI’s own model stack behind it.

First look at Grok Build by xAI, this time as a CLI tool running directly in the terminal. This is clearly positioned as a competitor to Claude Code, bringing agentic coding workflows into the command line with xAI’s own model stack behind it.

Mark Kretschmann

36,783 Aufrufe • vor 5 Monaten

Ornith 1.5 is a free MIT-licensed model that beats Claude Opus on some of the hardest tests, runs locally, works on your phone and teaches itself. Hit the top of Hacker News in hours.

Ornith 1.5 is a free MIT-licensed model that beats Claude Opus on some of the hardest tests, runs locally, works on your phone and teaches itself. Hit the top of Hacker News in hours.

0xMarioNawfal

43,901 Aufrufe • vor 2 Monaten

watch this anon. i gave NVIDIA's biggest model ever a single task. 100 minutes and 440,000 tokens later, it had rendered nothing. not one important thing on the screen. this is Nemotron 3 Ultra. 550 billion parameters, a hybrid Mamba Transformer MoE, the largest model NVIDIA has ever shipped, and they built it specifically for long-running agentic coding. so i handed it exactly that: build a 3D scene from a spec, multiple files, iterate until the tests pass. the same task a frontier model one shotted in minutes. i genuinely wanted to be impressed. it ran for an hour and forty. burned through 440,000 tokens. wrote every file, passed its own tests, and proudly printed "task complete."the browser was blank. the 3D scene never rendered. not once. and the long horizon agentic behavior was genuinely good. it stayed on task the whole hour and forty, wrote real multi-file code, drove its own tools without derailing. it just couldn't turn any of that into something that actually runs. here's the part that gets me. it's a text model, it cannot see its own output. so it sat there looping on a broken vision tool, trying to "look" at the page, hitting error after error, never once reasoning its way out. it declared victory on an empty screen because it had no way to know the screen was empty. to be fair, i genuinely don't know what quant the NIM was serving, so maybe some of that's on the serving, not the model. but the biggest model NVIDIA has ever made, on the exact task it was designed for, couldn't tell it had built nothing in 100 minutes. same task on a local model, below thread👇.

watch this anon. i gave NVIDIA's biggest model ever a single task. 100 minutes and 440,000 tokens later, it had rendered nothing. not one important thing on the screen. this is Nemotron 3 Ultra. 550 billion parameters, a hybrid Mamba Transformer MoE, the largest model NVIDIA has ever shipped, and they built it specifically for long-running agentic coding. so i handed it exactly that: build a 3D scene from a spec, multiple files, iterate until the tests pass. the same task a frontier model one shotted in minutes. i genuinely wanted to be impressed. it ran for an hour and forty. burned through 440,000 tokens. wrote every file, passed its own tests, and proudly printed "task complete."the browser was blank. the 3D scene never rendered. not once. and the long horizon agentic behavior was genuinely good. it stayed on task the whole hour and forty, wrote real multi-file code, drove its own tools without derailing. it just couldn't turn any of that into something that actually runs. here's the part that gets me. it's a text model, it cannot see its own output. so it sat there looping on a broken vision tool, trying to "look" at the page, hitting error after error, never once reasoning its way out. it declared victory on an empty screen because it had no way to know the screen was empty. to be fair, i genuinely don't know what quant the NIM was serving, so maybe some of that's on the serving, not the model. but the biggest model NVIDIA has ever made, on the exact task it was designed for, couldn't tell it had built nothing in 100 minutes. same task on a local model, below thread👇.

Sudo su

32,589 Aufrufe • vor 4 Monaten