I don't see the point in comparing yourself to Haiku which is not only useless for coding but also old. No thanks Microsoft.

"It is built end-to-end by Microsoft using clean and appropriately licensed data."

Well still no list nor publication of the training data.

>Build for developers, not benchmarks

That sounds like something you say when you don't benchmark well

Maybe this will replace raptor-mini as the "free" model on copilot plans? (but I don't see it at all yet on the student plan, in vscode or the cli)

How not to flex:

"MAI-Code-1-Flash outperforms Claude Haiku 4.5"

Raw feedback to the team: 1-model looks awesome, 2-The artificially smoothed scrolling on your page feels really bad!

I personally do not like Microsoft, but congrats them to release this model.

While the scores are not good compare to other open weight model, the important thing to note is their training data (as they claimed) is very clean, without any synthetic datasets.

Gemma 4 26B-A4B scored exceptionally well with 20% less params, so this isn't unprecedented.

The UI has Mustafa Suleyman written all over. Seems to be as much effort in rebranding MAI as in training.

Related ongoing thread:

MAI-Thinking-1 - https://news.ycombinator.com/item?id=48374362 - June 2026 (64 comments)