Reasonable

Building the verification layer for mission-critical software.

[Reasonable @ReasonableIO]

"The other half of the cure is proof."

[Azeem Azhar @azeem]

The internet’s Wuhan moment
https://t.co/iioL9iZoM7

[Reasonable @ReasonableIO]

Can open-weight models write challenging Verus proofs for TLA+ specs?

We tested @NVIDIAAI's Nemotron 3.5 Super VL and the answer is...absolutely, yes!

Our fine-tuning boosts performance from 43% to 73% on the hardest benchmark.

Read our latest report on novel datasets, fine-tuning 3.5 Super VL, and bridging the gap to real-world software verification:

https://t.co/lEGaxEWuLV

[NVIDIA AI @NVIDIAAI]

So many models launching this week!
And we have one of our own coming...⌛️

[Reasonable @ReasonableIO]

Formal verification of real-world software and systems requires models and toolkits capable across a wide variety of programming languages and verification techniques.

This focus on polyglot verification is one reason our translation and proof generation pipeline is so valuable. We also need proofs about a system's real-world behavior, which Verus (auto-active verifier built for Rust) enables.

All credit to Kseniia Strelbytska, Anna Mészáros, Balazs and the team working on this (amongst many other frontier-pushing initiatives)

-

Our paper “TLA-Verus: Generating and Solving Temporal Proof Obligations with LLM Agents,” has been accepted to the VeriCodeGen Workshop at NeurIPS 2026!

In brief: developers use TLA+ to design systems, describe their behaviour, and test them for flaws. But its main tool (TLC) only tests small versions of the system, and the design stays separate from the implementation code. We translate TLA+ designs into Verus, a verifier for Rust. There, LLM agents write proofs about the system’s behaviour for any run, at any size. Since the design now lives in Rust, it can later be linked to the actual code.

Our team generated 3,000+ fully machine-checked proofs of system properties, alongside a dataset to evaluate how well current LLMs write temporal proofs in Verus.

- Paper: https://t.co/ipGDXg9sgT
- New to TLA+? Read Reasonable’s intro: https://t.co/2KVkjzvHen
- Reasonable is hiring! https://t.co/pR0NnIkt4A

[Xudong Sun @xu_dong_sun]

I had a wonderful time with the Reasonable team discussing how Anvil's style of TLA proof in Verus is used and our new work Welder. Very grateful for the recognition and super excited to work on FV+AI together! Also love the T-shirts with my name on them :-) @fhuszar

[Reasonable @ReasonableIO]

Kseniia and Balázs are part of this cohort - brilliant impact with us already, setting the bar incredibly high.

You can read more about their work at Reasonable shortly

[Ferenc Huszár @fhuszar]

The level of mastery highschoolers reach in AI and computing today is astonishing!

This summer two AI Olympiad medalists, Kseniia and Balázs interned at @ReasonableIO, they both contributed incredible stuff pushing frontier of AI and formal reasoning, soaking up graduate level content before even starting their undergrad.

So glad to see this program embraced in the UK, run by the amazing @aalipie and James, supported by @matthewclifford, @KanishkaNarayan and many more.

[Reasonable @ReasonableIO]

We are hiring unusual CS, maths, machine learning and formal methods superstars:

🟥 shaping our research, engineering, and product development as our next Member of Technical Staff: https://t.co/0Tg1drMkSc

💙 driving growth-by-hiring as our Talent Partner: https://t.co/cQEhKgpGzt

🟡 building our operations in a tight, high-trust partnership as the EA to our founders: https://t.co/Dq7PLAGFEm

Play an early lead role in our growth - join to build and scale one of Europe’s most exciting AI startups! 🚀

[Reasonable @ReasonableIO]

Teach machines how to write provably correct code. On a boat. On the Danube. Join us: https://t.co/oqO8lLIbFR

[Craig Hughes 𓅪 @Craig_Hughes_]

Budapest remains wildly underrated. And the @ReasonableIO office boat on the Danube is a vibe

[Reasonable @ReasonableIO]

Ferenc is going to kill me for using the zee

CH

[Reasonable @ReasonableIO]

The future is specialized, token efficient, fast.

Great company to be in here too, thanks @NVIDIAAI

CH

[NVIDIA AI @NVIDIAAI]

Love seeing our partners post-training Nemotron 3.5 Lightning for their own domains, tools, and workflows, as well as those offering post-training support ⚡

Check out below to see what they’re building 🧵

[Reasonable @ReasonableIO]

Smaller, faster, capable: writing machine-checked proofs with a 30B open-weight model

We fine-tuned NVIDIA’s latest open model Nemotron 3.5 Lightning on 4B tokens of synthetic Verus data. The result: it beats a model ~50x its size on per-attempt pass rate while nearly matching it on pass@3, with faster token generation than any similarly sized open-weight model we tested.

Explore how the models we tested fail at formal proofs, how they cheat, and the effects of fine-tuning:
https://t.co/2hvlKhCUyK