Anthropic launched Claude Sonnet 5 on Tuesday, positioning the mannequin as a budget-friendly various to its flagship Opus line simply because the AI lab barrels towards a long-anticipated preliminary public providing.
The corporate is asking Sonnet 5 its “most agentic Sonnet mannequin but,” able to constructing plans, working browsers and terminals, and working autonomously at a degree that, till just lately, required far bigger and costlier fashions. Anthropic says the brand new launch considerably closes the efficiency hole with Opus 4.Eight whereas undercutting it sharply on value.
Benchmarks present a narrowing hole
On SWE-bench Professional, a extensively cited coding benchmark, Sonnet 5 scored 63.2%, effectively forward of predecessor Sonnet 4.6’s 58.1% and edging towards Opus 4.8’s 69.2%. On Terminal-Bench 2.1, the mannequin jumped to 80.4% from 67.0%, and tool-assisted reasoning climbed to 57.4% from 46.8%. In a knowledge-work benchmark, Sonnet 5 truly edged out Opus 4.8, scoring 1618 towards the bigger mannequin’s 1615.
“Opus 4.Eight remains to be the mannequin of selection for greater accuracy on these duties, however Sonnet 5 offers builders with lower-priced choices which might be of a lot greater high quality than what was beforehand accessible,” Anthropic stated in its announcement. The corporate added that customers can now toggle effort ranges between Sonnet 5 and Opus 4.Eight to steadiness value towards efficiency.
Sonnet 5 is near Opus 4.Eight ranges of intelligence, Supply: Anthropic
Pricing undercuts the sector
Sonnet 5 launches with introductory API pricing of $2 per million enter tokens and $10 per million output tokens, working by means of August 31, 2026, after which charges rise to $Three and $15 respectively. That continues to be effectively beneath Opus 4.8’s $5/$25 pricing, and Anthropic says it additionally beats OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Professional, although Google’s Gemini 3.5 Flash nonetheless is available in cheaper.
The mannequin is reside instantly throughout Free, Professional, Max, Crew, and Enterprise plans, the place it now serves because the default for Free and Professional customers. It’s additionally accessible in Claude Code and by way of the Claude API as claude-sonnet-5.
Early testers level to follow-through
A number of launch companions highlighted the mannequin’s tendency to complete multi-step jobs fairly than stall partway by means of. Daniel Shepard, a senior engineer at Zapier, stated the corporate handed Sonnet 5 a two-part job — updating Salesforce account tiers and sending a launch announcement to enterprise contacts — “and it completed finish to finish. That used to stall midway. For day-to-day automation, it’s a no brainer.”
Cursor co-founder Sualeh Asif stated the mannequin retains brokers “on plan,” following coding conventions by means of clear multi-step adjustments “at an environment friendly value.” Lovable co-founder Fabian Hedin praised its dealing with of unsafe requests, saying it “refuses unsafe requests cleanly and persistently.”
Security posture improves, however gaps with Opus stay
Anthropic says Sonnet 5 reveals a decrease general price of undesirable behaviors than Sonnet 4.6, together with lowered charges of hallucination and sycophancy, and is best at resisting prompt-injection hijack makes an attempt. The corporate has enabled real-time cyber safeguards by default, much like these used on Opus 4.7 and 4.8, although much less restrictive than the controls utilized to Fable 5.
On a Firefox exploit-development analysis run with Mozilla, neither Sonnet 5 nor Sonnet 4.6 produced a working exploit, although Sonnet 5 posted a better partial-success price (13.2% versus 8.8%). Each stay far behind Opus 4.8’s 68.8% and Mythos 5’s 88.4% on the identical check. Anthropic notes Sonnet 5 nonetheless reveals “considerably greater charges of misaligned conduct” than Opus 4.Eight and the corporate’s restricted Mythos Preview mannequin.
Context: an IPO push and a reshuffled mannequin lineup
The launch lands towards the backdrop of Anthropic’s confidential SEC submitting on June 1 for an IPO reportedly concentrating on a valuation close to $1 trillion. It additionally follows a turbulent stretch for the company’s top-tier lineup: Fable 5 and Mythos 5, Anthropic’s first Mythos-class fashions, launched June 9 earlier than a Commerce Division export-control directive pressured Anthropic to droop entry worldwide on June 12, citing considerations over a doable bypass of Fable 5’s cybersecurity safeguards.
Anthropic disputed the severity of the difficulty however complied, taking each fashions offline for each consumer globally because it couldn’t filter entry by nationality in actual time. On June 26, Commerce Secretary Howard Lutnick partially eased the order, permitting the unrestricted Mythos 5 for use by a slender set of vetted US critical-infrastructure organizations and authorities companies; Fable 5, the safeguarded model constructed for basic and industrial use, stays suspended for subscribers, API builders, and all worldwide prospects, with no restoration date introduced. OpenAI has faced a parallel intervention: the White House asked the company to limit the initial rollout of its GPT-5.6 lineup — Sol, Terra, and Luna — to a small group of government-approved companions after officers judged Sol’s capabilities “Mythos-like,” with OpenAI saying it expects broader availability “within the coming weeks” as each firms work with Washington on a longer-term evaluate framework for frontier fashions. Sonnet 5’s launch successfully continues enterprise as regular for Anthropic’s mid-tier lineup whereas the corporate’s strongest fashions stay caught in that unresolved regulatory standoff.
Jason Jones Jason Jones Read More
