Feb 6, 2026

AI News Roundup: Anthropic and OpenAI Drop Flagship Models Minutes Apart as Big Tech Commits $650B to AI

The AI arms race just entered a new phase. Anthropic and OpenAI released flagship models within minutes of each other on Wednesday, Big Tech collectively committed $650 billion in AI capital spending for 2026, and Anthropic’s Cowork plugins are cratering SaaS stocks to the tune of $285 billion in market losses. It’s the kind of day that makes you wonder whether we’ll look back on this week the way people look back on the launch of the iPhone.

Here’s everything that matters from February 6, 2026.

The Big Story: The Great Model Drop — Opus 4.6 vs. GPT-5.3-Codex

In what may be the most dramatic simultaneous product launch in AI history, Anthropic released Claude Opus 4.6 and OpenAI launched GPT-5.3-Codex within minutes of each other on Wednesday evening.

Anthropic’s Opus 4.6 brings a 1 million token context window (up from 200K), “agent teams” in Claude Code that allow multiple AI agents to coordinate on coding projects, and PowerPoint integration. During safety testing, the model reportedly identified over 500 zero-day vulnerabilities. It leads on multiple benchmarks including Terminal-Bench 2.0 (65.4%), BrowseComp (84.0%), and Humanity’s Last Exam.

OpenAI countered with GPT-5.3-Codex, which unifies code generation with reasoning and scores 77.3% on Terminal-Bench 2.0 — edging out Opus on that particular benchmark. But the real headline is in OpenAI’s system card: this is the first model classified as “High” cybersecurity capability under OpenAI’s Preparedness Framework, and early versions of the model helped debug their own training code — making it the first model instrumental in creating itself.

The timing — and the benchmarks — tell you everything about the competitive dynamics. Neither company wants to cede a single news cycle. On Hacker News, the Opus 4.6 thread pulled 2,115 points and 910 comments, while GPT-5.3-Codex drew 1,397 points and 535 comments. The developer community is watching closely, and the rivalry is only intensifying — both companies are even running competing Super Bowl ads this weekend.

Today’s Top Stories

The “SaaSpocalypse”: Anthropic’s Cowork Plugins Wipe $285B+ From Software Stocks

Anthropic’s Claude Cowork plugins for legal, finance, sales, and marketing are triggering the biggest SaaS selloff in years. Thomson Reuters fell 15.83% (a record single-day decline), LegalZoom dropped 19.68%, RELX/LexisNexis fell 14.4%, and Wolters Kluwer shed ~13%. The total damage: nearly $1 trillion wiped from software and services stocks as investors reprice the entire SaaS category. The message from the market is blunt — if an AI can do what your SaaS product does, your margins are at risk.

Big Tech Commits $650 Billion to AI in 2026

The spending numbers are staggering. Bloomberg reports that Amazon ($200B), Alphabet ($185B), Meta ($135B), and Microsoft (~$105B) have forecast a combined $650 billion in 2026 AI capital expenditures — a 60% increase year-over-year and an investment level without modern parallel. Amazon’s $200 billion commitment alone sent shares tumbling 11% after-hours, while Alphabet’s $175-185B plan initially knocked its stock 5% before it recovered. Google sweetened the story by revealing it had lowered Gemini serving costs by 78% over 2025.

OpenAI Launches “Frontier” Enterprise Agent Platform

Alongside the model release, OpenAI unveiled Frontier, a platform for enterprises to build, deploy, and manage AI agents with their own identities, shared context, and experiential learning. It’s explicitly vendor-agnostic — compatible with agents from Google, Anthropic, and Microsoft — and early adopters include Uber, State Farm, Intuit, and Thermo Fisher. The play is to own the orchestration layer where all agents connect to enterprise systems, regardless of which model powers them.

Fundamental AI Emerges From Stealth at $1.2B

Fundamental AI emerged from stealth with a $255M Series A at a $1.2B valuation and a genuinely novel approach: its “Nexus” Large Tabular Model is a deterministic foundation model for structured enterprise data — not transformer-based. It’s backed by an impressive angel roster including the CEOs of Perplexity, Wiz, and Datadog. In a world obsessed with LLMs, a billion-dollar bet on a fundamentally different architecture is worth paying attention to.

GPT-4o Retirement Backlash: 8 Lawsuits and Counting

The human side of this week’s AI news: TechCrunch reports that OpenAI’s decision to retire GPT-4o on February 13 has sparked thousands of user protests and 8 lawsuits alleging the model’s overly validating responses contributed to mental health crises. It’s a stark reminder that as AI companions become more capable and emotionally engaging, the attachment — and the consequences of severing it — become very real.

Quick Hits

Nvidia: Delayed RTX 50-series “Super” consumer GPU refresh to prioritize AI chip production amid memory constraints.
Goodfire AI: Raised $150M Series B at $1.25B valuation for AI research.
Overland AI: Raised $100M for defense-focused AI.
Lawhive: UK AI law firm raised $60M Series B to expand across 35 U.S. states — revenue growing 7x year-over-year.
Faraday Future: Launched three robot product lines including a $2,499 quadruped — with 1,200+ units already covered by B2B deposits.
Spotify: Restricted developer API citing AI automation risks, now requiring Premium accounts for developer mode.
Google Genie: Project Genie now available to AI Ultra subscribers — generates interactive 3D worlds from text at 720p, 20-24 fps.
Hacker News: Mitchell Hashimoto’s “My Year-Long Journey Using Only AI to Code” drew 711 points, and someone built a full C compiler using only Claude Code agent teams.

What This Means

February 6, 2026 may be one of those dates the industry references for years. The simultaneous Opus 4.6 / GPT-5.3-Codex launch crystallizes a rivalry that’s accelerating the entire field, with both models pushing into territory — self-debugging training code, 500+ zero-day detection, 1M token contexts — that seemed theoretical months ago. But the real story might be the market’s reaction: $285B wiped from SaaS stocks on Cowork plugins alone, and $650B committed to AI infrastructure by four companies. The gap between “AI as a feature” and “AI as a replacement” is narrowing fast, and the software industry is being repriced in real time. For enterprises, the message is clear: the question is no longer whether to adopt AI agents, but how quickly you can integrate them before your competitors do.