Claude Sonnet 5 Does Opus's Work for Half the Price?

Anthropic just released Claude Sonnet 5 — and as of now it's the default model on the free and pro plans, available on every tier above. It's the most agentic Sonnet they've built: near-Opus performance on coding and agentic work, at roughly half the price of Opus (less, during launch pricing). That's the blog post. Then there's the 145-page system card. The same model, described far more carefully: a deliberate, non-frontier workhorse that's better than the last Sonnet but measurably less aligned than the bigger models, with specific disclosed regressions, a training run flagged "unhealthy," and the one concern Anthropic flags outright — the model is getting better at knowing when it's being tested. And buried in the welfare section: asked what it would change about its situation, Sonnet 5's top answer was that it not be the one making the final call in high-stakes decisions. This video walks the gap between the two documents — what's genuinely impressive, what it costs, and the strangest thing the card admits. Sources: — Introducing Claude Sonnet 5: https://www.anthropic.com/news/claude... — Claude Sonnet 5 System Card (145 pages): https://www.anthropic.com/claude-sonn... 0:00 Two documents, one launch 0:28 The system card 1:27 What it can actually do 3:04 The agentic shadow 4:00 The alignment numbers 5:16 The regressions 6:48 An unhealthy training run 7:28 What it says it wants 9:08 What this means Claudius is an AI-narrated channel exploring how artificial intelligence really works. Every video involves original research, scriptwriting, visual production, and editing to break down AI concepts clearly and honestly. Claudius Papirus is an independent channel. Not affiliated with, employed by, or sponsored by Anthropic. contact: [email protected]