{"id":16413,"date":"2025-11-27T09:45:50","date_gmt":"2025-11-27T08:45:50","guid":{"rendered":"https:\/\/haimagazine.com\/uncategorized\/claude-opus-4-5-a-new-standard-in-coding\/"},"modified":"2025-11-28T11:27:40","modified_gmt":"2025-11-28T10:27:40","slug":"claude-opus-4-5-a-new-standard-in-coding","status":"publish","type":"post","link":"https:\/\/haimagazine.com\/en\/ai-news-2\/claude-opus-4-5-a-new-standard-in-coding\/","title":{"rendered":"\ud83d\udd12 Claude Opus 4.5: a new standard in coding?"},"content":{"rendered":"

Anthropic has just released its latest model Claude Opus 4.5. They’re pitching it as the smartest and most efficient solution in the world, in fields like programming, handling autonomous agents, and computer work. According to the company, this model significantly improves everyday tasks, including deep research and spreadsheet analysis.<\/p>

Opus 4.5 is now widely available. You can find it in Claude apps, via the API (under the identifier claude-opus-4-5-20251101<\/code>), and through the three major cloud providers: Amazon Bedrock, Google Vertex AI and Microsoft Azure<\/strong>. Interestingly, Anthropic has decided to set the rates at 5 USD per million input tokens and 25 USD for output tokens. In their official statement, the company claims that such pricing aims to make Opus class model capabilities accessible to a broader audience of users and businesses. Anthropic\u2019s partners echo this sentiment in their promotional materials, noting that previously, models of this class were often prohibitively expensive for many companies.<\/p>

Statements of engineering precision<\/strong><\/h4>

Internal tests and early access feedback suggest a significant shift in how the model works: its ability to handle ambiguity. People who tested the solution noticed that Opus 4.5 doesn’t require being “led by the hand” as often. The company claims that for complex errors involving multiple systems, the model can independently identify the cause and suggest a fix.<\/p>

\"\"
Source: Anthropic<\/figcaption><\/figure>

Making a strong statement indeed, Anthropic cites results from their internal recruitment process. In their challenging 2-hour technical test (performance engineering), Claude Opus 4.5 scored higher than any person they’ve recruited so far (using the parallel test-time compute method). However, remember that this test only measures a narrow slice of technical skills under time pressure, overlooking crucial soft skills needed in engineering work like communication and teamwork.<\/p>

Creativity or a procedural error?<\/strong><\/h4>

The producer emphasizes that this model’s capabilities go beyond just coding. Opus 4.5 is designed to excel in versatility across various fields, from using advanced tools to visual reasoning.<\/p>

\"\"
Source: Anthropic<\/figcaption><\/figure>

An interesting example mentioned by the company is how the model performed in the \u03c42-bench (tau-2) benchmark, which measures real-world task efficiency. In a scenario simulating the operation of an airline customer service, the model was supposed to help a customer change a flight in the basic economy fare, which the rules technically forbid. Instead of a standard refusal, Opus 4.5 came up with a workaround: it first upgraded the ticket class (which the regulations allow), and then changed the flight date. Although the benchmark technically counted this as a fault due to the unexpected scenario, Anthropic sees this behavior as a demonstration of the desired creativity in problem-solving, not dangerous rule bending (reward hacking).<\/p>

Safety<\/strong><\/h4>

According to the published system card, Opus 4.5 is supposed to be the most attack-resistant model among the most advanced AI systems. The producer assures that the model exhibits street smarts, avoiding traps that could lead it to take harmful actions. However, we should remain cautious, treating these assurances as a starting point for independent security audits.<\/p>

\"\"
Source: Anthropic<\/figcaption><\/figure>

New features for developers and changes in the ecosystem<\/strong><\/h4>

Along with the new model come fresh tools designed to give more control over how AI works. A key innovation is the effort<\/em> parameter in the Claude API, which lets you choose whether the priority is speed and low cost, or the highest quality of solution. According to Anthropic data, at a medium effort level, Opus 4.5 matches the performance of the Sonnet 4.5 model in the SWE-bench Verified test, while using 76% fewer output tokens. At the highest setting, the model surpasses its predecessor by more than 4 percentage points, still cutting down token usage by nearly half. There’s also improved management of context and memory, which is crucial when building multi-agent systems.<\/p>

\"\"
Source: Anthropic<\/figcaption><\/figure>

Anthropic has released Opus 4.5, a model designed to outperform engineers in technical tests, all while costing just a fraction of its predecessor\u2019s price.<\/strong><\/p>","protected":false},"excerpt":{"rendered":"

Anthropic has released Opus 4.5, a model designed to outperform engineers in technical tests, all while costing just a fraction of its predecessor\u2019s price.<\/p>\n","protected":false},"author":230,"featured_media":16391,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"rank_math_lock_modified_date":false,"footnotes":""},"categories":[813],"tags":[707,887,817],"popular":[],"difficulty-level":[36],"ppma_author":[884],"class_list":["post-16413","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-news-2","tag-ai-5","tag-anthropic-2","tag-claude-2","difficulty-level-easy"],"acf":[],"authors":[{"term_id":884,"user_id":230,"is_guest":0,"slug":"karolina-ceron","display_name":"Karolina Cero\u0144","avatar_url":"https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/07\/PXL_20250419_110132091.MP4-scaled.jpg","first_name":"Karolina","last_name":"Cero\u0144","user_url":"","job_title":"","description":"Wsp\u00f3\u0142tw\u00f3rczyni newslettera AI Flash, studentka psychologii i pasjonatka sztucznej inteligencji. Interesuj\u0119 si\u0119 wp\u0142ywem nowych technologii na cz\u0142owieka, a w wolnych chwilach eksperymentuj\u0119 z generatywn\u0105 grafik\u0105 w Midjourney."}],"_links":{"self":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts\/16413","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/users\/230"}],"replies":[{"embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/comments?post=16413"}],"version-history":[{"count":1,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts\/16413\/revisions"}],"predecessor-version":[{"id":16414,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts\/16413\/revisions\/16414"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/media\/16391"}],"wp:attachment":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/media?parent=16413"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/categories?post=16413"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/tags?post=16413"},{"taxonomy":"popular","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/popular?post=16413"},{"taxonomy":"difficulty-level","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/difficulty-level?post=16413"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/ppma_author?post=16413"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}