Quixotic AI

AI sovereignty for the world's most trusted runtime.

[Quixotic AI @qxoticai]

AI on the JVM: Quixotic?

How fast can an LLM written in pure Java really be? Faster than C++?

All the answers today, 11:50, Room 4. #Devoxx

https://t.co/YdEAl2B4Lc

[Quixotic AI @qxoticai]

Works on any machine with a JVM. No GPU needed.
Run it yourself, one command (first run downloads and caches the model, ~14GB).

[Quixotic AI @qxoticai]

Spring Boot shops, banks, insurers, payroll systems, all the backends that quietly run the world: millions of JVM apps can now add a 24B-A2B reasoning model as a plain Maven dependency. Every Liquid AI model is supported.
No infra. No ops. No Python.

1009 tok/s prefill (~2.3x llama.cpp)
47.8 tok/s decode (parity)

https://t.co/XcCU9cw3PN

[Liquid AI @liquidai]

Today we release 𝙻𝙵𝙼𝟸.𝟻-𝟸𝟺𝙱-𝙰𝟸𝙱, the largest 𝙻𝙵𝙼𝟸.𝟻 family member. It is an efficient local model for laptops and PCs to work through multi-step tasks inside popular agent harnesses.

> 24B total parameters, 2B active per token
> Context expanded to 131K tokens
> A 128K-token vocabulary
> 33T pretraining tokens
> It is a reasoning model
> Very easy to fine-tune and customize
> Performs competitively with Qwen3.6-35B-A3B, a model with ~50% more total and active parameters, on agentic and tool-use tasks.
> Open base and post-trained weights on @huggingface 🤗

1/

[Quixotic AI @qxoticai]

@codepreneur247 @java @Android The pragmatic path to run it on Android would be to compile jinfer to a shared native library with GraalVM's Native Image e.g. https://t.co/wi4UPnyjlw
Jinfer heavily uses MemorySegment ... a rewrite to use ByteBuffer instead would be possible too.

[Stephan Janssen ☕️🧠🧞‍♂️ @Stephan007]

@qxoticai The first #Devoxx talk I watched offline! 🤩

Super interesting presentation and amazing work! ☕️👏 #Congrats

This could be a great fit for real-time transcripts at #Devoxx Belgium 2027.... Also excited to explore how we could integrate it with @DevoxxGenie 🚀

[Quixotic AI @qxoticai]

The video is out!
My #Devoxx talk, "Quixotic: Pure Java Inference at Native Speed" is now available:
https://t.co/1zzZNDviHx

[Quixotic AI @qxoticai]

AI in a jar!

Source code: https://t.co/SBNmElAD0R

Project page, benchmarks and examples: https://t.co/8S2e267X5H

#Devoxx #Java

[Quixotic AI @qxoticai]

The talk was just the trailer.
Benchmarks: https://t.co/8S2e268uVf
Source: https://t.co/SBNmElBaQp
#Devoxx #Java

[Quixotic AI @qxoticai]

@dordokamaisu @alina_yurenko @TheMukel @Devoxx Recording is out!
https://t.co/1zzZNDviHx

[Quixotic AI @qxoticai]

AI inference engine for the JVM.
https://t.co/8S2e267X5H

[Quixotic AI @qxoticai]

Finally, AI in a jar!
#Devoxx

[Quixotic AI @qxoticai]

@ynorsk @alina_yurenko @thomaswue @TheMukel Yes!
The next target will be consumer accelerators with shared memory e.g. Apple Silicon and AMD X Halo series.

[Quixotic AI @qxoticai]

AI sovereignty for the world's most trusted runtime: the JVM.
#Devox

[Аlina Yurenko 🇺🇦 @alina_yurenko]

GraalVM CE compiler confidently beating C++ in inference workloads 🚀🚀🚀
By @TheMukel at #Devoxx

[Quixotic AI @qxoticai]

AI sovereignty for the world's most trusted runtime: the JVM.
#Devox

[Quixotic AI @qxoticai]

Benchmarks and examples at:

https://t.co/2BcvSe4Vnh

[Quixotic AI @qxoticai]

AI on the JVM: Quixotic?

How fast can an LLM written in pure Java really be? Faster than C++?

All the answers today, 11:50, Room 4. #Devoxx

https://t.co/YdEAl2B4Lc

[Quixotic AI @qxoticai]

AI on the JVM: Quixotic?

How fast can an LLM written in pure Java really be? Faster than C++?

All the answers today, 11:50, Room 4. #Devoxx

https://t.co/YdEAl2B4Lc

[Quixotic AI @qxoticai]

Our pure Java matmul implementation using the Vector API is competitive with llama.cpp's.
https://t.co/TrQ9xEaUWh

Right now, the absolute top performance for quantized matrix multiplication requires specialized VNNI instructions. The other kernels run pretty fast in pure Java using the Vector API.

[Quixotic AI @qxoticai]

AI in Java... quixotic, until now!

End-to-end AI stack for the JVM, built from first principles: Tokenizers, quantized matrix multiplication, model format readers, inference engine, vision, speech: all in Java. No external services, no wrappers, no Python, no ONNX.

Will be presented at Devoxx Belgium, next week: "Quixotic: Pure Java Inference at Native Speed"
https://t.co/CS91m6Yzb8

[Quixotic AI @qxoticai]

@RecentlyBaned C++? C'est tout Rust maintenant. Voici ta pilule rouge: https://t.co/2BcvSe5tcP

[Quixotic AI @qxoticai]

"Write once, run anywhere!"
Not "Write once, bridge to Python!"
https://t.co/O3kZU89tL5