Anthropic ships Claude Sonnet 5 as agentic capability becomes the mid-tier floor

Near-Opus performance at $2/$10 introductory tokens, and a mandate to finish the task without being asked twice.

Anthropic released Claude Sonnet 5 on Tuesday at introductory API rates of $2 per million input tokens and $10 per million output, roughly 60% below Opus 4.8’s $5/$25, and posted benchmark scores that sit within a few points of the flagship. On SWE-bench Pro, Sonnet 5 lands at 63.2% against Opus 4.8’s 69.2% and Sonnet 4.6’s 58.1%. On Terminal-Bench 2.1, it’s 80.4% versus 82.7%. On Humanity’s Last Exam with tools, 57.4% versus 57.9%. On the GDPval-AA v2 knowledge-work benchmark, Sonnet 5 actually edges Opus, 1,618 to 1,615.

The framing is that agentic capability is now baseline at every price tier. TechCrunch, noting OpenAI’s GPT-5.6 Sol preview last week and Google’s Gemini 3.5 Flash launch in May, argues the differentiator is no longer whether a model can run agents but how cheaply and reliably. Sonnet 5 is Anthropic’s answer: default on Free and Pro, available to Max, Team, Enterprise, in Claude Code, and via the claude-sonnet-5 API.

Read the launch copy carefully and the product thesis is behavioral, not architectural. Anthropic’s early-access testers describe a model that “finishes complex tasks where previous Sonnet models would stop short” and “checks its own output without explicitly being asked.” Lovable co-founder Fabian Hedin told TechCrunch it “refuses unsafe requests cleanly and consistently.” Anthropic also concedes Sonnet 5 has “a much lower ability to perform cybersecurity tasks” than current Opus models, a safety-posture disclosure that doubles as market segmentation.

Two footnotes matter for anyone modeling spend. Introductory pricing expires August 31, 2026, when rates rise to $3 and $15. And the new tokenizer can map the same text to up to 1.35x more tokens than the previous one; Anthropic says pricing was set to keep the switch roughly cost-neutral, but VentureBeat notes the expansion is a variable enterprises should model themselves before trusting the headline rate.

The industry has stopped selling capability and started selling cost per completed task. That’s the whole story.

Anthropic ships Claude Sonnet 5 as agentic capability becomes the mid-tier floor

Sources