Microsoft’s 7 MAI Models, Anthropic’s IPO Filing, and the Week AI Got Serious

This week the AI industry didn’t just move fast — it redrew the map. Microsoft went full vertical with seven in-house models, Anthropic quietly filed for what could be the biggest IPO in tech history, and GitHub Copilot fundamentally changed how developers pay for AI assistance. Let’s break it all down.

Microsoft Build 2026: The “Hill-Climbing Machine”

The headline from Microsoft Build 2026 isn’t just the models — it’s the process. Microsoft AI CEO Mustafa Suleyman framed the entire MAI family as a “hill-climbing machine”: an integrated system of data pipelines, training infrastructure, and iterative self-improvement designed to push the frontier continuously, not just ship checkpoints.

The seven models:

• MAI-Thinking-1 — Microsoft’s first reasoning model. Medium-sized, but hits 97.0% on AIME 2025 and 94.5% on AIME 2026. That’s competitive with the best reasoning models from OpenAI and Anthropic.
• MAI-Code-1-Flash — A lean coding model baked directly into GitHub Copilot and VS Code. This is Microsoft’s answer to the “we rely too much on OpenAI” problem.
• MAI-Voice-2 and MAI-Voice-2 Flash — Next-gen speech generation models.
• MAI-Transcribe-1.5 — Multilingual speech recognition.
• MAI-Image-1 and MAI-Video-1 — First-party image and video generation.

The strategic signal is unmistakable: Microsoft is building its own AI stack from scratch. After years of leaning on OpenAI’s GPT models, the MAI family represents a clear pivot toward model independence. As the Verge coverage put it, this is about “lessening reliance on OpenAI and lowering costs.”

Anthropic Files for IPO: The $1 Trillion Question

On June 1st, Anthropic confidentially submitted a draft S-1 to the SEC, officially entering the IPO race alongside OpenAI and SpaceX. The company had just closed a $65 billion funding round at a $965 billion valuation — and now they’re eyeing public markets.

A few things make this filing significant:

• First-mover advantage: Anthropic beat OpenAI to the confidential filing, giving them a potential edge in timing.
• Profitability narrative: The company told investors it expects its first-ever quarterly operating profit in Q2 2026. In a sector burning cash at unprecedented rates, that matters.
• Open-source pressure: Unlike OpenAI, Anthropic has maintained a more research-transparent posture, which could play well with public market investors who want to see a path beyond API revenue.

The Bloomberg take is that Anthropic’s lead over OpenAI in the IPO race is set to widen. Whether that translates to a better valuation depends on what numbers are hiding in that confidential filing.

GitHub Copilot’s Billing Revolution

As of June 1st, GitHub Copilot moved to usage-based billing powered by GitHub AI Credits. The old flat-rate Premium Request Units are gone. Now your bill depends on which model you use and how many tokens you burn.

What changed:

The developer backlash has been real — Reddit threads are full of people doing the math and finding their costs going up. But GitHub’s argument is that this aligns cost with actual value: if you’re using a cheaper model for autocomplete and only hitting Opus for complex reviews, you should pay proportionally.

For teams running Copilot at scale, budget management just became a first-class concern. Expect more tooling around cost monitoring to emerge in the next few months.

The Coding Agent Wars: Claude Code vs Cursor vs Everyone Else

The agent landscape is consolidating around a few key players. Recent benchmarks and real-world usage data paint a clear picture:

• Claude Code leads on reasoning quality and large-codebase work. It’s the terminal-native choice for engineers who want deep, autonomous agents.
• Cursor dominates UX and file-aware editing. It’s where you go for interactive, fast-paced development with the best autocomplete in the business.
• Cline remains the open-source default for VS Code users who want step-by-step approval and transparency.
• MAI-Code-1-Flash is Microsoft’s new entry, and it’s free inside Copilot — which could be a serious disruptor.

The Morph rankings note that Claude Code alone accounts for 135K GitHub commits per day. That’s not a tool — that’s an ecosystem.

What’s interesting is the emerging pattern: developers aren’t picking one agent, they’re composing them. Claude Code for complex features, Cursor for daily editing, Copilot for quick completions. The “best agent” question is becoming “best agent stack.”

The Bigger Picture

Three threads converge this week:

1. Model ownership matters. Microsoft building its own models isn’t just a cost play — it’s about controlling the entire stack from silicon to IDE.
2. AI is going public. Anthropic’s IPO filing signals that the AI industry is mature enough (and profitable enough) to face public market scrutiny. That changes everything about how these companies operate.
3. The unit economics of AI assistance are being rewritten. GitHub’s billing shift means developers need to think about AI costs the way they think about cloud costs — as something to monitor, optimize, and budget for.

The spaghetti is getting tangled, but the signal is clear: 2026 is the year AI infrastructure becomes real business, not just demos and developer previews.

This post contains affiliate links. If you sign up for services through links in this post, we may earn a commission at no extra cost to you. Disclosure

Want this in your inbox every morning? Subscribe to the Spaghetti Stories newsletter.