In development · Coming soon
Nimbus v3.
A little storm
is taking shape.
The next generation of Nimbus local models is being built for coding, reasoning and tool use. Your work stays on your machine. Our focus stays on making more of it possible.
Local by design.
Memory-conscious.
Measured before release.
One family. Different footprints.
A place on your machine.
Four planned sizes, from compact everyday use to more demanding local work. Preview artwork below; final names, formats and availability will follow evaluation.
What matters next
Smaller is useful.
Capable is essential.
We’re exploring quantized downloads and vocabulary compression alongside coding, language and tool-use evaluations. A smaller file only earns its place if the model remains useful.
- Memory, measured.
- Download size, runtime memory and speed will be measured separately—not treated as interchangeable numbers.
- Two ways to work.
- Direct answers and native thinking are being evaluated separately, including how the models choose and use tools.
- Results before claims.
- Comparisons with Nimbus v2 come before release claims. No benchmark scores or memory savings are being announced here.
Your code. Your machine.
Good things take a little weather.
No release date has been announced. Follow the current Nimbus collection for released models, or keep using v2 while the next family takes shape.
Explore current releases