Anthropic introduced Claude Sonnet 5 on June 30 — and this is more than a point update. According to Anthropic, Sonnet 5 is the most agentic Sonnet model yet: it makes plans, uses tools like browsers and terminals, and runs autonomously at a level that would have required larger and more expensive models just a few months ago.
Close to Opus, at half the price
The exciting part is the price-performance ratio. On many tasks, Sonnet 5 comes close to Opus 4.8 — but costs much less. At launch there’s an introductory price of 2 dollars per million input tokens and 10 dollars per million output tokens, valid through August 31, 2026. After that it’s 3 and 15 dollars respectively. For comparison: Opus 4.8 sits at 5 and 25 dollars.
That makes Sonnet 5 cheaper not just than Anthropic’s own flagship, but reportedly also than OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro. If you run agents at scale, you feel that directly on the bill.
What stands out in practice
Anthropic shows benchmarks against Sonnet 4.6 and Opus 4.8. But the real win is in the behavior: early testers describe Sonnet 5 finishing complex tasks where older Sonnet models would stop short — and checking its own output without being explicitly asked.
One tester put it nicely: Sonnet 5 is at its best on brownfield code — race conditions, hidden tests, the parts nobody wants to touch. Instead of patching the symptom, it traces a failure to its actual root cause and ships a durable fix. Another handed it a bug to investigate: unprompted, it wrote a reproducing test, implemented the fix, then stashed it to confirm the bug came back without the change — all in a single pass.
Availability and safety
Sonnet 5 is available across all plans starting today. It’s the default model for Free and Pro, and available to Max, Team, and Enterprise as well. It runs in Claude Code and on the Claude Platform as claude-sonnet-5. A new effort control lets you fine-tune the balance between cost and performance.
On safety, Anthropic reports an overall lower rate of undesirable behavior than Sonnet 4.6 — fewer hallucinations, less sycophancy, better resistance to prompt injection. Cyber capabilities sit well below the Opus models, yet Sonnet 5 still launches with real-time cyber safeguards enabled.
My take
The trend is clear: it’s not the most expensive model that wins, but the one with the best ratio of capability to price. Sonnet 5 shifts exactly that boundary. For most agent workflows it’s likely the default from today — Opus stays for the genuinely hard cases. And the timing, right alongside Claude Science and a day before the Fable release, shows just how tight Anthropic is running its cadence right now.
Sources: Anthropic: Introducing Claude Sonnet 5, TechCrunch, VentureBeat, Simon Willison