Here are the 5 biggest AI stories from the week of April 13–19, 2026.

1. Anthropic: Claude Opus 4.7

Anthropic released Claude Opus 4.7 this week. The headline numbers:

  • 3× more production bugs fixed on SWE-bench compared to the previous model
  • 98.5% on vision tasks — near-perfect image understanding
  • New /ultrareview command in Claude Code for deeper code analysis

The catch: Opus 4.7 uses a new tokenizer that adds up to 35% more tokens on code-heavy tasks, which increases cost even though the per-token price stayed the same. If you work with a lot of code, run the math on your current usage before upgrading.

2. OpenAI: GPT-5.4-Cyber

OpenAI launched GPT-5.4-Cyber, its first model built specifically for defensive security work.

Key capabilities:

  • Reverse-engineers binaries without access to source code
  • Analyzes malware and suspicious code at a level previous models couldn’t
  • Tiered access — you need to apply through OpenAI’s Trusted Access for Cyber program to get in

This is not a public release. It’s aimed at security researchers and vetted organizations. But it signals that OpenAI is taking the security tooling market seriously.

3. MiniMax: M2.7 Open-Sourced

Chinese AI lab MiniMax open-sourced M2.7, a 229B parameter model.

What stands out:

  • The model was partially used in its own training pipeline — it helped train itself
  • 56% on SWE-Pro, a demanding software engineering benchmark
  • Available for free on Hugging Face, and compatible with Ollama for local inference

A 229B model that runs on Ollama is a big deal for anyone who wants a capable local model. It’s not lightweight, but if you have the hardware, it’s worth testing.

4. OpenAI: London Office + Stargate UK

Two conflicting OpenAI UK news items dropped in the same week:

  • OpenAI opened its first permanent London office, signaling long-term commitment to the UK market
  • Stargate UK, the planned £100B AI infrastructure project, was paused due to a funding review

The pause doesn’t mean the project is cancelled, but it does mean the timeline is unclear. The contrast between opening a permanent office and pausing a flagship infrastructure program in the same week raised a lot of questions.

5. PwC Study: The AI Gains Gap

A PwC study this week found that just 20% of companies are capturing 75% of AI’s economic gains.

The dividing line is not industry or company size — it’s how deeply AI is integrated into core workflows. Companies that treat AI as a productivity tool for individual workers are falling behind. Companies that rebuild processes around AI are pulling ahead.

The gap is getting wider, not narrower.

Watch the Short

I break down all 5 stories in under 90 seconds on YouTube: