The Coding Agent War Is Now a Three-Front Battle — and OpenAI Is Playing Defense

OpenAI raised $122B and launched Codex everywhere, but Anthropic's limited Mythos release and Claude's desktop takeover tell a different story about where the real leverage is shifting.

The week's real story isn't about who raised the most money

OpenAI just closed a staggering $122 billion funding round — the largest private tech financing in history. AI companies collectively hauled in $297 billion in Q1 alone, per the New York Times. And yet, the most interesting signal this week came from Anthropic restricting access to its new model, not expanding it.

Let's unpack what's actually happening across the big vendors.

OpenAI: Spending Big, Stretching Thin

OpenAI pushed Codex — its agentic coding tool — aggressively this week, launching a VS Code extension, a ChatGPT integration, and even a Claude Code plugin (yes, really — codex-plugin-cc lets you use Codex from inside Claude Code). They also rolled out a new $100/month ChatGPT Pro tier with 5x Codex usage limits.

But the cracks are visible. The Verge reports that "the vibes are off at OpenAI," with the company sitting at an $852 billion valuation while its own CFO told The Information she's unsure revenue will support spending commitments. Sora, their AI video tool, shut down after just six months. And Axios highlighted a telling cultural difference: Anthropic emphasizes efficiency in training and inference, while OpenAI's mindset has been that Sam Altman can always raise more capital.

App store rankings tell their own story: ChatGPT leads downloads, but Gemini is second and Claude third. xAI's Grok has dropped to 22nd place on Apple's App Store.

Anthropic: The Efficiency Play

Anthropic made two moves that deserve close attention. First, Claude Code Desktop now includes computer use — AI that can control your mouse, keyboard, and screen to complete tasks. It launched on Windows this week after a macOS preview. Alongside it, Claude Cowork positions Claude as an autonomous knowledge worker that operates in your local files and applications, not just a chat window.

Second, and more provocative: Anthropic announced Mythos, a new model it explicitly won't release publicly. Per the New York Times, Anthropic described Mythos's ability to find software security vulnerabilities as a "cybersecurity reckoning" — and limited access to a handful of tech companies. This is a deliberate signal: we built something dangerous, and we're choosing restraint over revenue.

Whether you find that admirable or theatrical, it's a clear differentiation strategy against OpenAI's "ship everything" approach.

The Rest of the Field

Google is quietly building out Gemini Code Assist with MCP support — that's the Model Context Protocol, an open standard that lets AI tools connect to external data sources and services. Mistral secured $830 million in debt financing to build a GPU-packed data center in Paris, betting that European sovereignty over AI infrastructure will matter. Meta continues expanding the Llama ecosystem, now available through AWS Bedrock, Ollama, and seemingly everywhere else.

And a wildcard: Jeff Bezos's Project Prometheus is reportedly poaching top engineers from both OpenAI and xAI, per Observer.

What to Watch

For business leaders: The coding agent war matters because it determines which vendor controls your engineering workflow. Codex, Claude Code, and Gemini Code Assist are all competing to become the default layer between your team and your codebase. Evaluate now — switching costs will compound.

For builders: MCP adoption is accelerating across vendors, including OpenAI. If you're building AI integrations, MCP is becoming the USB-C of AI connectivity — learn it.

For everyone: Watch the Anthropic efficiency narrative versus OpenAI's spending narrative. One of these strategies is wrong, and we'll know which by year-end.

The bottom line: The vendor with the best model no longer wins automatically — the vendor that embeds deepest into your daily workflow does, and that race just got a lot more interesting.

Sources