Finn | Something new is coming...
Joined March 2024
- Tweets 186
- Following 11
- Followers 91
- Likes 492
9 Photos and videos
Meet FinnOS (In development)
> Fully Agentic OS
> Not Linux, not Android, our own Custom Microkernel built on seL4 microkernel
> Uses programming languages and toolkits adapted for AI Agents
> Built for the era of AGI
> Fully OSS under GPL-2.0 License
github.com/Finn-Technologies…
2
1
3
241
e1 is almost AGI.
Got the question wrong in the end but it did consider the car being really dirty.
e1 is our first AI model.
We fine-tuned Gemma 4 QAT E4B, the results left us impressed.
We were able to achieve very advanced reasoning and Chain of Thoughts, while also being 3x more accurate at predicting the next tokens.
Try it and be shocked.
huggingface.co/Finn-Technolo…
1
1
8
We’d have liked to add benchmarks.
Sadly we don’t have a great computer, the training happened on an M4 MacBook Air with 16GB RAM and limited storage.
We can’t run traditional benchmarks like SWE-Bench because those require more power.
We’d love to see you try the model, tho.
e1 is our first AI model.
We fine-tuned Gemma 4 QAT E4B, the results left us impressed.
We were able to achieve very advanced reasoning and Chain of Thoughts, while also being 3x more accurate at predicting the next tokens.
Try it and be shocked.
huggingface.co/Finn-Technolo…
1
1
24
e1 is our first AI model.
We fine-tuned Gemma 4 QAT E4B, the results left us impressed.
We were able to achieve very advanced reasoning and Chain of Thoughts, while also being 3x more accurate at predicting the next tokens.
Try it and be shocked.
huggingface.co/Finn-Technolo…
2
2
5
613
Finn retweeted
Jun 15
What Lisa Su actually held on stage:
A mini PC the size of a lunchbox running Qwen3-235B locally, with no cloud and no discrete GPU
Inside: the Ryzen AI Max 395, 128GB unified memory, 110GB usable as VRAM on Linux
The first x86 chip that handles 200 billion parameters on a single die
AMD claims it beats the RTX 5080 by several times on memory-bound models — because the 5080 simply cannot fit them
$1,400 to $2,500 once. cloud bills run $200 to $400 a month
It pays for itself in a few months, then costs nothing per request
This is not a faster GPU. it is the first real argument that your AI does not belong in someone else's data center
Community note
Price is incorrect. It's going for about USD$3,999. See: microcenter.com/search/search_… and amd.com/en/products/pr…
207
419
3,834
1,148,460
We’re introducing almost the same thing but with Gemma 4 QAT E4B
Jun 14
This is the most hilarious thing I saw and did today
Ran gemma-4-12B-coder-fable5-composer2.5-v1-GGUF locally with 8 GB VRAM at 20 tok/sec
Anthropic's Claude Fable 5 launched June 9.
By June 12 it was banned. I can't access it. You can't either.
But here's the twist: I'm running a model trained on its chain of thought at 20 tok/s on my RTX 4060 8GB.
Locally. Offline. No cloud. No export control.
Enter: Gemma4-12B-Coder GGUF (Q4_K_M)
Base: Google's gemma-4-12B-it
Fine-tuned on verifiable Python CoT data:
- Primary: Composer 2.5 real reasoning traces (only passing solutions kept)
- Auxiliary: Fable 5 used to redo the hard cases Composer missed.
Every training example's reasoning led to code that actually ran. No hallucinated logic.
Llama.cpp flags:
-m gemma4-coding-Q4_K_M.gguf -cnv -ngl 44 -c 64000 -v
(huggingface model link in comments)
Flag breakdown:
-ngl 44 → offload 44 layers to GPU (tune this for your VRAM)
-c 64000 → 64K context window
-cnv → conversation/chat mode
-v → verbose output
The irony writes itself.
Anthropic spent weeks telling the world Fable 5 (mythos) is too powerful to release. Then released it. Then got banned from serving it, including their own researchers.
Meanwhile: a Gemma 4 12B fine tune, trained on Fable 5's reasoning, runs fully offline on my mid range consumer GPU
No API. No cloud. Just me and llama.cpp.
This is why local AI matters.
Check out the model's link in the comments. How's your experience been with this model?
1
2
90
The OS has Agentic AI built into it,
Let’s say you want to make an app:
- You prompt
- Agent for code and testing spawn
- Both work in parallel, one codes and one tests
- You keep using your machine, while the 2 agents have built your app, added features, and it’s ready to ship.
2
43
We’re excited as a new member (@Waleedkafafi) joins the team!
Expect Finn Software to become even better, and keep an eye for upcoming Flux v0.2 👀
1
1
4
296
Flux with similar features will require:
- 4GB RAM
- Snapdragon 695/Dimensity 6300 or Snapdragon 765G and equivalent and later
- OS at minimum Android 10, Windows 10 and macOS 11 Big Sur
Everything runs completely offline, you’re welcome ;)
Spec requirements for Gemini Intelligence on Android devices 🫠
1
2
167
People want this level of integration local?
Then we’ll make it local.
Google just showed everyone what an “AI-native laptop” actually means, and it’s not what you think.
Everyone’s waiting for the magic chip that makes AI free and local.
-
But nothing Google showed says that’s actually happening.
You’re most likely still paying for Gemini, and the heavy lifting is still living in the cloud.
The important part is AI finally escaped the chatbot tab:
- It’s in your cursor.
- It generates widgets on your desktop.
- It can pull from Files, Gmail, Calendar, and your phone as context.
It’s not a tool you go to anymore. It’s becoming part of how the OS works.
Now it’s Apple’s turn to answer.
1
2
108
Flux v0.1.9 has released!
This update includes:
- Gemma 4 E2B and E4B replacing Qwen 3.5 2B and 4B
- Image processing
- Faster response generation
Flux is now the fastest Local AI app out there!
You can download Flux here:
github.com/Finn-Technologies…
1
1
2
124