Microsoft unveils seven new AI models pitched on price-performance against GPT-6 rivals
Summarized from the Hacker News discussion because the article couldn't be read
- A commenter quotes the page saying Decision 1.1 is 50x faster and 79x cheaper than GPT-6.1 Sol and is used by 4,000 customers.
- A reply from a self-identified member of the Microsoft team says MD1.1 scores 68 on the Artificial Analysis index and costs $140 per 1M tasks, versus $156 for Jev and $345 for Gemini.
- A commenter says MAI Code 2 Flash is priced at 65% of GPT-6 Luna and is marketed for coding only.
- Commenters say the open-weight question matters most, with one saying they would pay Microsoft to host open-weight models but not use closed ones.
Hacker News opinions
MAI Code 2 Flash at 65% of GPT-6 Luna's price is a nice number, but it's pitched as coding only. I'd like it for general stuff too. Locally hosted Gemma4 already beats the cheap Sonnet and Luna options for me on language tasks.
50x faster and 79x cheaper than GPT-6.1 Sol is a useless comparison. Nobody picks a model off that line.
They're selling to executives, not devs. Microsoft has always been good at selling upstairs, and devs picking the tool is probably over.
The '#x times cheaper' phrasing is annoying. Just give me a percentage.
Someone from the team says MD1.1 scores 68 on Artificial Analysis and costs $140 per 1M tasks vs Jev's $156. Per-token pricing matches Jev, so the gap is fewer tokens per task.
Open weights or I'm not interested. I'd pay Microsoft to host them, but I won't touch closed models when the market is already full of them.
The Windows Phone comparison doesn't hold up. Phones are sticky, AI plans aren't. Teams swap providers every few months now.
Plenty of companies only use Copilot because they already have Microsoft contracts. Nobody wants to start a new relationship with a lab.
Microsoft would rather let something die than open source it. That's just how they are.