Evals, RL environments & AI data | Building benchmarks for binary reasoning.
Austin, TX
Joined December 2009
- Tweets51.1K
- Following1.4K
- Followers1.4K
- Likes6.8K
Meet Mistral Large 4, aka Le Chonk.
• 1T parameters, natively multimodal. 49B active.
It is the best open weights model from US or Europe on aggregated benchmarks.
• State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding.
• Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure.
• Available to all via API today. Working with cybersecurity partners privately.
Open weights release end of October.
Every coding agent I’ve used has the same issue: it forgets your repo as soon as the session ends. Come back six months later and it’s still grepping around trying to figure out the same stuff it already learned.
Tether exist so it doesn't have to.
It remembers things like who owns what, what tends to break, and patterns you’ve already decided not to use. And it ties that memory back to actual files and commits, so it can check whether something is still true.
You can switch models without starting from zero.
So far, we’re seeing roughly a third fewer tool calls on the same questions, with better answers. If you're on mac download the GUI.
Local-first and open source:
github.com/agentrebench/Teth…
