![\"\"](\"https:\/\/www.europesays.com\/ai\/wp-content\/uploads\/2026\/06\/sonnet_5_benchmarks-scaled-1.webp\")
<\/a>Sonnet 5 beats its predecessor, Sonnet 4.6, across every tested category and closes in on the pricier Opus 4.8. On knowledge work (GDPval-AA v2), Sonnet 5 even edges past Opus 4.8 with 1,618 points versus 1,615. | Image: Anthropic<\/p>\nOn the knowledge work benchmark GDPval-AA v2,\u00a0which tests AI on real-world knowledge tasks<\/a>, Sonnet 5 actually beats the larger Opus 4.8, scoring 1,618 to Opus’s 1,615. Anthropic says feedback from early-access partners told the same story. Sonnet 5 acts far more agentically than previous versions, showing up in things like how it handles search tasks.<\/p>\n Lately, Anthropic has been making news for models it\u00a0can’t\u00a0ship. The\u00a0US government is blocking the company’s two most capable models, Mythos 5 and Fable 5<\/a>, over cybersecurity concerns. That context hangs over the Sonnet 5 launch. Anthropic is clearly eager to get ahead of any similar worries. The model wasn’t trained on cybersecurity tasks, the company says, and in tests for risky capabilities like writing software exploits, it scores far below both Opus 4.8 and Mythos 5.<\/p>\n Sonnet 5 does score a bit higher than its predecessor on these tasks, though. So Anthropic has switched on\u00a0cyber safeguards<\/a>\u00a0by default. They flag and block risky cyber usage in real time, on par with the protections already in place for Claude Opus 4.7 and 4.8. They’re dialed back compared to Fable 5’s guardrails, which users\u00a0complained about almost immediately<\/a>. Anthropic says it views the overall cybersecurity risk from Sonnet 5 as low.<\/p>\n On the safety front, the model does a better job turning down malicious requests and fending off prompt injection attacks than Sonnet 4.6, according to Anthropic. Hallucinations and\u00a0sycophantic behavior<\/a>, the tendency to just agree with whatever the user says, are down as well. Anthropic’s full safety evaluation is in the\u00a0Claude Sonnet 5 System Card<\/a>.<\/p>\n Introductory pricing runs through August 2026<\/p>\n Claude Sonnet 5 is live now on all plans. It’s the new default for Free and Pro users, and Max, Team, and Enterprise subscribers can access it too. Developers can plug it into Claude Code and the Claude Platform. On the API side, it goes by\u00a0“claude-sonnet-5”<\/a>. The training cutoff is January 2026, with a one-million-token context window.<\/p>\n Until August 31, 2026, Anthropic is charging $2 per million input tokens and $10 per million output tokens.\u00a0After that<\/a>, prices jump to $3 and $15, which is what previous Sonnet models cost.<\/p>\n Real-world costs might tell a different story:\u00a0Because the model works more agentically, it’s likely to chew through more tokens per task<\/a>. So even at the same per-token rate, running Sonnet 5 could end up costing more than its predecessors. The same thing happened when Opus went from 4.6 to 4.7.<\/p>\n \t\t\t\tAI News Without the Hype \u2013 Curated by Humans<\/p>\n \n\t\t\t\t\tSubscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive “AI Radar” frontier report six times a year, full archive access, and access to our comment section.\t\t\t\t<\/p>\n<\/a>Agentic search performance on BrowseComp by effort level and cost per task. Sonnet 5 (orange) clearly outperforms Sonnet 4.6 (gray) at every level while offering cheaper entry points. Opus 4.8 (yellow) stays ahead at the highest effort settings. | Image: Anthropic
\nCybersecurity isn’t a concern this time<\/p>\nFirefox 147 exploit evaluation. Like its predecessor Sonnet 4.6, Sonnet 5 couldn’t develop a fully working exploit but shows a slightly higher partial control rate at 13.2 percent. Mythos 5 and Opus 4.8 are far more capable at this task. | Image: Anthropic<\/p>\n