Reading model benchmarks like a pro, Mythos is looming, and Claude talk caveman, save big token

About this episode

The episode explores a new Claude plugin that reduces AI output to primitive shorthand and discusses Anthropic's efforts in cybersecurity and recent open-source model releases.

Is the secret to slashing your token costs by 65% forcing your LLM to speak like a caveman? This week on the Friday Deploy, Andrew and Ben test out a hilarious new Claude plugin that reduces AI output to primitive shorthand before diving into Anthropic's $100 million push to win the cybersecurity arms race with Project Glasswing. The hosts also unpack the sudden release of four game-changing open-source models—including Gemma 4 and Holo3—and explain why modern AI benchmarks are proving that h...

People in this episode

Hosts: Andrew, Ben

Topics covered

• AI benchmarks
• token costs
• cybersecurity
• open-source models
• Claude plugin
• primitive shorthand

Keywords

• token costs
• AI
• Claude
• cybersecurity
• open-source
• Gemma 4
• Holo3
• Project Glasswing

Mentioned in this episode

Organizations: Anthropic

Products: Claude, Project Glasswing, Gemma 4, Holo3

More episodes of Dev Interrupted

• Microsoft breaks free from OpenAI, using your harness to add drag instead of velocity, and the Linux built-ins you're sleeping on · June 5, 2026 · 27 min
• How to turn your 1000x engineer into a 10x everyone | LinkedIn’s Karthik Ramgopal · June 2, 2026 · 53 min
• The cost of intelligence will never be this cheap again, the failure of intensive specs, and how bots disguise inefficient workflows · May 29, 2026 · 39 min
• Observability is your profit center now | Honeycomb’s Christine Yen · May 26, 2026 · 48 min
• What Google didn’t announce at I/O, defining “dark flow” and ignoring your first brain to build your second one · May 22, 2026 · 30 min
• Android is the frontier for agents and other lessons from Google I/O | Matthew McCullough · May 19, 2026 · 27 min