OpenAI and Anthropic drop models minutes apart

OpenAI and Anthropic drop models minutes apart

From The Bit Baker newsletter — February 7, 2026

PLUS: $650B AI spending rattles Wall Street, and a viral AI agent goes rogue

Good morning, Dave. Anthropic and OpenAI just staged the most dramatic product launch in AI history — releasing competing frontier models within 15 minutes of each other on Wednesday, February 5, 2026.

Opus 4.6 brings a million-token context window and autonomous agent teams, while GPT-5.3-Codex counters with self-debugging capabilities and OpenAI's first-ever "High" cybersecurity rating. When the two biggest AI labs are racing to the minute, what does that mean for the rest of the industry — and for your stack?

In today's Bit Baker:

• Anthropic and OpenAI's dueling frontier model drops
• OpenAI's new Frontier platform for enterprise AI agents
• Big tech's $650B AI spending rattles Wall Street
• A viral AI agent's security crisis sounds the alarm

Anthropic and OpenAI drop dueling AI models minutes apart

The Bit Baker: In a dramatic display of one-upmanship, Anthropic released Claude Opus 4.6 and OpenAI launched GPT-5.3-Codex within 15 minutes of each other on Wednesday — both pushing hard into agentic coding and enterprise AI.

Unpacked:

• Opus 4.6 introduces a 1-million token context window in beta, multi-agent "agent teams" that spin up parallel workers autonomously, and adaptive thinking that adjusts reasoning depth across four effort levels.
• GPT-5.3-Codex is 25% faster than its predecessor, partly debugged its own code during training, and is the first OpenAI model rated "High capability" for cybersecurity — prompting a new trusted-access program for security professionals.
• Both models set new benchmarks for agentic coding tasks, though early testers noted the improvements feel incremental rather than a generational leap over their predecessors.

Bottom line: The simultaneous launches signal that the AI arms race between Anthropic and OpenAI is now measured in minutes, not months. Developers get two powerful new tools — but the real story is how fast the frontier is moving.

OpenAI launches Frontier — its platform for enterprise AI agents

The Bit Baker: OpenAI introduced Frontier, a new platform that lets enterprises build, deploy, and manage AI agents like human employees — complete with onboarding, performance reviews, and scoped permissions.

Unpacked:

• Frontier connects agents to existing business systems like CRMs and data warehouses, giving them shared institutional memory so they can plan, execute tasks in parallel, and handle complex workflows across departments.
• The platform includes enterprise-grade governance with agent IAM, scoped permissions, and compliance with SOC 2 Type II and ISO 27001 — a direct answer to companies worried about giving AI agents access to sensitive data.
• Early adopters include HP, Intuit, Oracle, State Farm, and Uber, with use cases spanning financial forecasting, customer support, and regulatory workflows in life sciences.

Bottom line: Frontier marks OpenAI's shift from model provider to enterprise infrastructure company. It's a bet that the next wave of AI adoption won't come from chatbots — it'll come from AI agents embedded in business operations.

Big tech's $650B AI spending bet wipes $1.35 trillion from stocks

The Bit Baker: Tech stocks lost over $1.35 trillion in a single week as investors panicked over the staggering scale of AI infrastructure spending planned for 2026 — a combined $660 billion from Amazon, Meta, Alphabet, and Microsoft.

Unpacked:

• Amazon's $200 billion capex forecast — a 56% year-over-year increase and the highest among hyperscalers — shocked analysts, triggering an 8% single-day stock drop and dragging Microsoft, Nvidia, and Oracle down with it.
• The total planned AI infrastructure spend now exceeds the combined GDP of the UAE, Singapore, and Israel, raising questions about whether enterprise AI adoption can justify the buildout.
• Morningstar called the situation "becoming binary" — either the spending delivers a massive payoff or it becomes a huge waste of shareholder cash, with no middle ground in sight.

Bottom line: The market is sending a clear message: believing in AI's potential is one thing, but $660 billion in a single year is testing even the most optimistic investors' patience. Expect these spending debates to define tech stocks for the rest of 2026.

OpenClaw's viral AI agent spawns a massive security crisis

The Bit Baker: OpenClaw, the open-source AI agent that exploded to 150,000+ GitHub stars in weeks, has become the first major security disaster of the agentic AI era — with researchers finding nearly 12% of its marketplace skills are malicious.

Unpacked:

• Security researchers audited all 2,857 skills on ClawHub and found 341 malicious ones, with 335 linked to a coordinated campaign called "ClawHavoc" — a single actor published 314+ trojanized skills disguised as crypto wallets, YouTube tools, and auto-updaters.
• The deeper problem is indirect prompt injection: attackers embed hidden instructions in emails, Slack messages, and Google Docs that OpenClaw's agent executes using valid OAuth tokens, turning everyday collaboration tools into attack vectors.
• Shodan scans found 24,478 internet-exposed OpenClaw instances, with at least two belonging to U.S. organizations valued over $20 billion — and one in five companies deployed it without IT approval.

Bottom line: OpenClaw is a preview of what happens when AI agents get deep system access before security catches up. As the industry races to ship autonomous agents, this is the cautionary tale every developer and IT team needs to read.

The Shortlist

Mistral released Voxtral Transcribe 2, a family of on-device speech-to-text models — including a 4B-parameter open-weights model under Apache 2.0 with sub-200ms latency and support for 13 languages at $0.003 per minute.

Positron raised $230M in Series B funding at a $1B+ valuation for its energy-efficient AI inference chips, with an upcoming Asimov chip promising 5x more tokens per watt than Nvidia's Rubin GPU.

Amazon announced a closed beta launching in March for AI-powered film and TV production tools at MGM Studios, aiming to help produce 14-20 theatrical films in 2026 — up from 5-8 previously.

Reddit reported Q4 ad revenue of $690M — a 75% year-over-year jump driven by AI-powered search, targeting tools, and its new Reddit Max campaigns that delivered 27% more conversions.

References

1. HAL 9000 — Wikipedia
2. Claude Opus 4.6 — Anthropic
3. Introducing GPT-5.3-Codex — OpenAI
4. GPT-5.3 Codex warns of unprecedented cybersecurity risks — Fortune
5. Two new models — Simon Willison
6. Introducing OpenAI Frontier — OpenAI
7. OpenAI launches a way for enterprises to build and manage AI agents — TechCrunch
8. Big tech loses $1.35T as AI spending fears spark selloff — TechBuzz
9. Data center capex spending: $630 billion — Fortune
10. AI leaders argue: software adapt or die, valuations stretched — Fox Business
11. The OpenClaw security saga — Cyera
12. From automation to infection: how malicious skills spread — VirusTotal
13. Voxtral Transcribe 2 — Mistral
14. Positron raises $230M Series B to take on Nvidia's AI chips — TechCrunch
15. Amazon to begin testing AI tools for film and TV production — TechCrunch
16. Reddit advertising surges in Q4, grows user base — MediaPost