Daily AI CatchupClaude Sonnet 5 Launches With Near-Opus Performance at Half the Cost
Anthropic released Claude Sonnet 5, calling it the most agentic Sonnet model yet — capable of operating a browser or terminal and sustaining longer multi-step jobs, with knowledge-work capability reportedly surpassing Opus 4.8. It's priced at $2/$10 per million input/output tokens through August 31 (then $3/$15), though cybersecurity benchmarks regressed versus Sonnet 4.6, a side effect of Anthropic avoiding those training tasks during the Fable export saga.
Why it matters
💻 Developer · Sonnet 5 can drive a browser or terminal directly and holds up over longer agentic runs — worth benchmarking against whatever model you're currently using for multi-step tool-calling tasks.
📦 Product · Near-Opus knowledge-work quality at Sonnet pricing changes the cost math for any feature you'd previously reserved for your most expensive model tier.
🎨 Design · Longer-horizon agentic runs mean AI can now carry a design or content task through more steps unattended before needing a check-in.
📈 Business · Introductory pricing ($2/$10 per million tokens) holds through August 31 — a window worth locking in volume commitments before the price step-up to $3/$15.
🤔 Just Curious · Anthropic's new mid-tier model, Sonnet 5, now performs close to their most expensive model but costs a fraction as much — and it can control a computer's browser and terminal on its own.
Try this: Compare Sonnet 5 against your current default model on a real multi-step agentic task (e.g., research-then-summarize) before the introductory pricing expires August 31.
Sources: Anthropic: Claude Sonnet 5