Microsoft Goes Full-Stack: 7 In-House AI Models, Anthropic Eyes $1T IPO, and the Coding Agent Wars Heat Up

The AI industry doesn’t take weekends off. While the markets were closed, the past few days have been absolutely stacked with moves that reshape the competitive landscape. Microsoft just told OpenAI “we can do this ourselves,” Anthropic is eyeing a trillion-dollar IPO, and GitHub Copilot’s billing change is hitting developers’ wallets right now.

Let’s break it all down.

Microsoft’s Build 2026 Bombshell: Meet the MAI Family

The biggest story of the week — and possibly the month — is Microsoft unveiling seven in-house AI models at Build 2026. This isn’t just a side project. This is Microsoft AI CEO Mustafa Suleyman going all-in on self-sufficiency.

The lineup:

• MAI-Thinking-1 — Microsoft’s first reasoning model, a 35B active parameter beast trained from scratch with zero distillation. It’s positioned to go head-to-head with Claude Opus and GPT-5.5 on complex reasoning, math, and general intelligence tasks.
• MAI-Code-1-Flash — An inference-efficient agentic coding model, already integrated into GitHub Copilot and Visual Studio Code. This is Microsoft’s answer to Claude Code and Codex, running on their own silicon.
• MAI-Scout — An always-on agent model for persistent, lightweight tasks.
• MAI-Vision-1, MAI-Voice-1, MAI-Image-1 — Multimodal models covering the full spectrum.
• MAI-Majorana-2 — A specialized model (details still emerging).

The message is clear: Microsoft is done being just an OpenAI reseller. They’re building a full-stack AI empire, from reasoning to coding to multimodal, all under the MAI (Microsoft AI) brand.

What’s smart about MAI-Code-1-Flash is the integration angle. It’s not a separate product — it’s baked into the tools millions of developers already use. That’s a distribution advantage that indie coding agents can only dream of.

Anthropic Files for IPO: The $1T Question

While Microsoft was dropping models, Anthropic confidentially filed for an IPO that could value the company at up to $1.77 trillion — which would make it the largest IPO in history.

Let that sink in. A company that didn’t exist 5 years ago could debut at nearly double Microsoft’s current market cap.

The numbers behind the filing:

• $65 billion raised in late May at a $965 billion post-money valuation — already ahead of OpenAI
• Pricing guidance of $135/share would put the company at $1.77T
• Expected to join SpaceX and OpenAI as the three trillion-dollar listings of 2026

And they’re not just riding hype. Claude Opus 4.8 dropped on May 28th, fixing the tool-calling and comment-verbosity issues from 4.7 while delivering benchmark gains at the same price point ($5/$25 per million tokens). Anthropic’s differentiation narrative is sharp: lead on coding agents and enterprise safety, let OpenAI fight the consumer crowd.

The timing is interesting too. Going public while your flagship model is on a hot streak and your coding agent (Claude Code) is widely considered the best in class? That’s not desperation — that’s maximizing leverage.

GitHub Copilot’s Billing Revolution: AI Credits Are Here

As of June 1, 2026, GitHub Copilot moved from request-based billing to usage-based billing with GitHub AI Credits. If you’re a Copilot user, this is already affecting you.

What changed:

• Premium Request Units (PRUs) are gone, replaced by GitHub AI Credits
• Cost now depends on which model you use and how many tokens you consume
• Code completions remain unlimited on all plans
• Copilot Chat and cloud agents consume credits at model API rates
• Copilot code review now consumes GitHub Actions minutes on top of AI credits
• User-level budgets and a “Copilot Max” upgrade path are available

The developer backlash has been real. The old PRU system was predictable — you knew what you were getting. Now, a long Copilot cloud agent session burning through GPT-5.5 tokens can eat credits fast. For teams running autonomous agents at scale, this could get expensive quickly.

The silver lining? The shift to usage-based pricing means lighter tasks cost less. Quick chat questions with lightweight models might cost a fraction of a credit. It’s the cloud computing model all over again — pay for what you use, but watch your usage.

The Coding Agent Wars: A Fractured Battlefield

The coding agent landscape in 2026 is a messy, competitive, and fast-moving space. Here’s where things stand based on the latest comparisons and real-world testing:

The Tier 1 contenders:

The rising challengers:

• Microsoft MAI-Code-1-Flash — Native Copilot integration is a massive distribution play
• Aider — Git-native, terminal-first, great for pair programming
• Antigravity — New entrant getting buzz for speed
• Kilo Code — VS Code-native agent gaining traction

The trend is clear: every major AI company wants to own the coding agent layer. It’s the most visible, most measurable, and most immediately valuable application of AI right now. Microsoft building their own model specifically for this? That tells you everything.

Trump’s AI Executive Order: Voluntary Oversight

On June 2nd, President Trump signed an Executive Order on AI in the National Security Enterprise, establishing a voluntary framework where AI developers can give the government up to 30 days to review advanced models before public release.

Key points:

• Voluntary, not mandatory — companies choose to participate
• Focused on cybersecurity and national security applications
• Google, Microsoft, and xAI already agreed to share early model access with the government
• OpenAI and Anthropic had previously reached similar agreements with the Pentagon

The CFR analysis calls it a signal that the administration is shifting from pure deregulation toward light-touch oversight. Whether this actually constrains frontier AI development or just creates a review bottleneck remains to be seen.

The Bigger Picture

This week’s moves reveal three macro trends:

1. Vertical integration is accelerating. Microsoft building its own models, Anthropic going public to fund independence — nobody wants to depend on anyone else’s API.

2. The coding agent is the new browser. It’s the interface through which developers interact with AI, and every major player is fighting to own that layer. The model matters less than the workflow integration.

3. Pricing models are being stress-tested. GitHub’s shift to usage-based billing, Anthropic holding prices steady on Opus 4.8, Microsoft bundling MAI-Code-1-Flash into existing Copilot plans — the economics of AI coding tools are still being figured out.

The indie/open-source coding agent community faces an existential question: can you compete when Microsoft, Anthropic, OpenAI, and Google are all building agents with the full weight of their platforms behind them?

The answer, for now, is yes — because open-source agents like Cline offer something the big players can’t: model agnosticism, local execution, and zero vendor lock-in. But the window might be narrowing.

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

Want this in your inbox every morning? Subscribe to the Spaghetti Stories newsletter for daily AI dispatches.