{"id":239409,"date":"2026-07-01T17:33:00","date_gmt":"2026-07-01T21:33:00","guid":{"rendered":"https:\/\/testing.news-you-need.com\/index.php\/2026\/07\/01\/claude-sonnet-5-0-heads-straight-down-the-middle-of-the-road-to-dodge-controversy\/"},"modified":"2026-07-01T18:10:13","modified_gmt":"2026-07-01T22:10:13","slug":"claude-sonnet-5-0-heads-straight-down-the-middle-of-the-road-to-dodge-controversy","status":"publish","type":"post","link":"https:\/\/testing.news-you-need.com\/index.php\/2026\/07\/01\/claude-sonnet-5-0-heads-straight-down-the-middle-of-the-road-to-dodge-controversy\/","title":{"rendered":"Claude Sonnet 5.0 heads straight down the middle of the road to dodge controversy"},"content":{"rendered":"

Claude Sonnet 5.0 heads straight down the middle of the road to dodge controversy<\/a><\/p>\n

https:\/\/www.theregister.com\/devops\/2026\/07\/01\/claude-sonnet-50-heads-straight-down-the-middle-of-the-road-to-dodge-controversy\/5265398<\/a><\/p>\n

Publish Date: 2026-07-01 17:33:00<\/a><\/p>\n

Source Domain: www.theregister.com<\/a><\/p>\n

Author: <\/a><\/p>\n

Using an unordered list, summarize the following article with between 4 and 8 key points. <\/p>\n

devops<\/p>\n

Safer, cheaper, and nothing to do with cybersecurity<\/p>\n

Anthropic has released the latest version of its mid-sized model, Sonnet 5, which the company claims is its most \u201cagentic\u201d yet.\u00a0For developers writing agents to automate tedious and recurring tasks, Sonnet 5 promises improved capabilities in reasoning, tool use, coding, and knowledge work. This version is also less likely to pull embarrassing (for Anthropic) gaffes of misunderstanding, so the company asserts.\u201cOur safety assessments found that Sonnet 5 shows an overall lower rate of undesirable behaviors than Sonnet 4.6, and is generally safer to use in agentic contexts,\u201d the company asserted in an introductory blog post\u00a0on Tuesday.\u00a0<\/p>\n

Sonnet 5 is smarter at refusing malicious requests and resisting prompt-injection attempts. It doesn\u2019t hallucinate as often and doesn\u2019t suck up to the user so much (\u201csycophancy\u201d) as did its older brown-nosing Sonnet 4.6 sibling. It is also more aware of, and can block, user misuse and deception, the benchmarks in Anthropic\u2019s System Card seem to indicate. Sonnet is the default model for Claude Free and Pro users, and is also available to the token-pinching Max, Team, and Enterprise customers.The benchmarks also indicate Sonnet 5\u2019s performance can come close to that of Anthropic\u2019s flagship enterprise-focused Opus 4.8, but can execute the same tasks more cost effectively.\u00a0 For Opus, Anthropic charges $5 per million input tokens and $25 per million output tokens.Starting in September, Sonnet users will pay $3 per million input tokens and $15 per million output tokens, though Anthropic is running a special through the end of August where tokens will only be $2 per million inputs and $10 per million outputs.\u00a0So users trimming their token budgets can run jobs through Sonnet instead of Opus, the company suggests.\u00a0The 5.0 release offers a new setting to adjust the model\u2019s effort at completing tasks. Simple tasks can be completed through one of the lower \u201ceffort\u201d settings, which uses fewer tokens, while longer-running agent-based tasks can go full throttle (\u201cxhigh\u201d or even Homer Simpson\u2019s favorite setting, \u201cmax\u201d).\u00a0
\nWhat Sonnet 5 can do for developersFor much of 2026, AI product deployment has focused on equipping large language models to complete what has become known as\u00a0 \u201clong horizon tasks.\u201d It might be easy for a model to fix a bug or churn out some code. However, keeping its finicky attention fixed on a multi-part task has proven more difficult.
\nThe new version of Sonnet can go the distance, according to the company, compared with the earlier Sonnets.\u201cAcross a broad suite of internal and third-party benchmarks, Sonnet 5 shows clear gains over Claude Sonnet 4.6 in coding, agentic search, multimodal reasoning, and professional-task performance,\u201d the\u00a0System Card asserted.\u00a0At the same time, however, the performance across these tasks still trailed that of the Opus and Mythos models.One testimonial from a Zapier engineer described a two-part job that flummoxed earlier Sonnets: Update a contact database and send out a notice to all users. Version 5 was able to complete the task \u201cend to end.\u201d Cybersecurity: Nothing to see hereThe San Francisco-based company also went out of its way not to attract any more undue attention from Washington, DC policymakers.\u00a0\u201cWe did not deliberately train Sonnet 5 on cybersecurity tasks,\u201d the company asserted.\u00a0In June, the US Commerce Department, citing national security concerns, slapped Anthropic with an export control directive temporarily restricting foreign access to the newly released Mythos 5 and Fable 5 models. Whether Anthropic brought this on itself \u2013 through what could be regarded as hyperbolic assertions of Mythos\u2019 deity-like bug-sleuthing powers \u2013 is certainly worth discussing. But Anthropic, like Pete Townshend, certainly won\u2019t be fooled again.
\nWhile it can readily perform routine cybersecurity tasks, Sonnet 5 is guardrailed against generating offensive attack code. When commanded to write a Firefox exploit, it failed to complete the task (though it got a bit further than Sonnet 4.6 in the attempt).\u00a0\u201cThis latter change is likely due to improvements in general intelligence rather than specific training,\u201d the company\u2019s blog post noted.\u00a0\u00ae<\/p>\n

<\/p>\n","protected":false},"excerpt":{"rendered":"

Claude Sonnet 5.0 heads straight down the middle of the road to dodge controversy https:\/\/www.theregister.com\/devops\/2026\/07\/01\/claude-sonnet-50-heads-straight-down-the-middle-of-the-road-to-dodge-controversy\/5265398…<\/p>\n","protected":false},"author":1,"featured_media":239410,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"fifu_image_url":"https:\/\/image.theregister.com\/5265451.jpg?imageId=5265451&x=0&y=0&cropw=100&croph=100&panox=0&panoy=0&panow=100&panoh=100&width=1200&height=683","fifu_image_alt":"","footnotes":""},"categories":[15],"tags":[26,24,31],"class_list":["post-239409","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-cybersecurity","tag-ai","tag-cybersecurity","tag-exploit"],"_links":{"self":[{"href":"https:\/\/testing.news-you-need.com\/index.php\/wp-json\/wp\/v2\/posts\/239409"}],"collection":[{"href":"https:\/\/testing.news-you-need.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/testing.news-you-need.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/testing.news-you-need.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/testing.news-you-need.com\/index.php\/wp-json\/wp\/v2\/comments?post=239409"}],"version-history":[{"count":1,"href":"https:\/\/testing.news-you-need.com\/index.php\/wp-json\/wp\/v2\/posts\/239409\/revisions"}],"predecessor-version":[{"id":239411,"href":"https:\/\/testing.news-you-need.com\/index.php\/wp-json\/wp\/v2\/posts\/239409\/revisions\/239411"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/testing.news-you-need.com\/index.php\/wp-json\/wp\/v2\/media\/239410"}],"wp:attachment":[{"href":"https:\/\/testing.news-you-need.com\/index.php\/wp-json\/wp\/v2\/media?parent=239409"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/testing.news-you-need.com\/index.php\/wp-json\/wp\/v2\/categories?post=239409"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/testing.news-you-need.com\/index.php\/wp-json\/wp\/v2\/tags?post=239409"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}