Microsoft's MAI announces in-house SOTA pre-trained base models in 3T-parameter weight class
tatsu_hashimoto · x · 2026-10-24
Mustafa Suleyman announced that MAI has built state-of-the-art pre-trained base models in the 3T-parameter weight class, trained entirely in-house at Microsoft, with a technical blog post detailing the work and post-training results promised soon.
In a quoted follow-up, an instructor noted his CS 312 students independently discovered related rules in an assignment, comparing their results against MAI's released tutorials.
More from Companies & People
- Two core Claude Code creators announce departures from Anthropic the same day — DrDatta_AIIMS · 2026-10-24
- Ari Emanuel on why he did a five-day meditation retreat at his peak: dyslexia, identity, fear — davidsenra · 2026-10-24
- How Bridgewater turned 100M documents into an edge with its Vault knowledge base — AI Engineer · 2026-10-24
- Without Kimi, Z.ai and DeepSeek, OpenAI and Anthropic might have struck a secret 'détente' — doodlestein · 2026-10-24
- Scholé AI CEO to present the learning science behind multi-agent adaptive tutors, plus Berkeley/EPFL RCT — Cohere_Labs · 2026-10-24
- Cohere Labs hosts Oct 28 research pitches on RL, LLM reasoning and robot learning — Cohere_Labs · 2026-10-24