What shipped today<\/h2>\n
Per Anthropic’s announcement<\/a>, Claude Opus 5 is live now on Claude.ai, in Claude Code and Claude Cowork, and on the API under the model id Anthropic describes it as a “thoughtful and proactive model that comes close to the frontier intelligence of Fable 5 at half the price.” Bloomberg’s framing<\/a> was blunter: a more cost-efficient model for everyday work. Both are accurate. This is a workhorse release, and workhorse releases are the ones that actually change what your monthly bill buys.<\/p>\n API pricing is unchanged from Opus 4.8: $5 per million input tokens and $25 per million output tokens (Anthropic<\/a>). Two dials sit on top of that base price:<\/p>\n The price-performance math is the real story. Anthropic says<\/a> Opus 5 lands within 0.5% of Fable 5’s peak score on CursorBench 3.2 at half the cost, and beats Fable 5 on OSWorld 2.0 computer-use tasks at one third the cost. When a model this close to the frontier costs half as much, “which model can I afford” quietly stops being the question for most teams.<\/p>\n Put numbers on it. A moderately busy agent workload burning 50 million input and 10 million output tokens a month runs about $500 on Opus 5. The same workload on Fable 5 pricing costs roughly double, and the benchmark gap you are paying that premium for has narrowed to low single digits on the coding evaluations Anthropic published today. For a solo operator or a small team, that difference is a real line item, and it compounds every month the workload runs.<\/p>\n Opus 4.8, which shipped in May at this same price point, was already the everyday flagship for serious agentic work. We covered it in our Opus 4.8 breakdown<\/a>, and today’s release supersedes much of that advice for new projects. Per Anthropic<\/a>, the concrete gains over 4.8:<\/p>\n Two smaller ships matter for anyone building on the API, per 9to5Mac<\/a>: you can now change which tools Claude uses mid-conversation without invalidating your prompt cache, and requests flagged by safety classifiers fall back to an alternative model instead of erroring out. The second one removes a class of 3 a.m. pipeline failures that had nothing to do with your code.<\/p>\n Agents.<\/strong> Self-verification is the trait to watch. The expensive part of running agents is the human time spent reviewing output and re-prompting when the model confidently ships a flaw. A model that audits itself before returning, at Opus pricing, moves more workflows from “supervised” to “spot-checked.” If you are setting up your first agent stack, start with our agent setup guide<\/a> and slot Opus 5 in as the default worker model.<\/p>\n Coding.<\/strong> State of the art on Frontier-Bench and near-Fable performance on CursorBench means solo developers and small teams get close to frontier code quality at half the frontier bill. The effort dial fits here too: high effort for architecture and gnarly debugging, low effort for boilerplate and refactors.<\/p>\n Content operations.<\/strong> Better judgment, fewer unforced errors, and stronger visual output shorten the edit pass, which is where content costs actually live. Our own posts run through an engine built on models in this class, with a human editor on every piece. If you want that setup without building it yourself, The Blogging System<\/a> starts at $25\/mo or $197\/yr, includes 5 drafts a month, carries a 30-day money-back guarantee, and the list you build is yours to export. This blog is drafted with Empower Network’s AI content engine and edited by a human before publishing.<\/em><\/p>\n Non-developers.<\/strong> You do not need an API key to feel this release. Opus 5 is already the default on Claude Max and the top model on Claude Pro (9to5Mac<\/a>), and it runs inside Claude Cowork for document and spreadsheet work. The practical difference on a subscription is fewer rounds of “no, fix the second column” per task. Anthropic’s claim that the model requires less back-and-forth is the one subscription users can test today against their own most annoying recurring task, no benchmark required.<\/p>\n Research-heavy work.<\/strong> Fortune<\/a> reports Anthropic is positioning Opus 5 as its most capable generally available model for scientific research, and naming law firms and software teams as target users. If your work is reading dense material and producing defensible analysis, this tier just got cheaper.<\/p>\n Opus 5 is not the top of Anthropic’s line. It trails Fable 5 on some tasks, and Anthropic itself still recommends Fable 5 for advanced autonomous projects (Fortune<\/a>). It also remains behind Mythos 5 on cybersecurity and autonomous biology evaluations, areas Anthropic gates deliberately (Anthropic<\/a>).<\/p>\n Benchmark caveats apply as always. Frontier-Bench v0.1 and CursorBench 3.2 are young evaluations, vendor-reported numbers favor the vendor, and independent replication takes weeks. Treat today’s figures as Anthropic’s claims until third parties confirm them. We will append corrections here if the independent numbers diverge.<\/p>\n Who can ignore this release: anyone whose workload already runs fine on Sonnet 5. If your agents browse, fill forms, and draft at the quality you need for $2\/$10 intro pricing, upgrading buys you capability you are not using. Same for casual chat users; on free and Pro plans the difference shows up mostly in long, multi-step work, not quick questions. And if you standardized on Opus 4.8 last month, nothing breaks today. The same money simply buys more when you migrate, so migrate on your schedule, not Anthropic’s.<\/p>\n Here is the current Anthropic ladder for operators, as of today. Sonnet 5 is the volume tier: cheap, fast, good enough for most routine agent traffic. Opus 5 is now the default work tier: near-frontier quality for coding, analysis, and agents at $5\/$25, with effort levels to tune spend per task. Fable 5 stays the escalation tier for the hardest autonomous projects, at roughly double the cost; our Fable 5 explainer<\/a> covers when that premium pays for itself.<\/p>\n A sensible pattern: route everything to Sonnet 5 first, escalate to Opus 5 when quality or task length demands it, and reserve Fable 5 for the small slice of work where being slightly wrong is expensive. Migration itself is low-drama since pricing did not move: swap the model id, rerun whatever evaluation set you trust on your own tasks, and check the results before flipping production traffic. If you have no evaluation set, this release is a good excuse to write down five representative tasks and grade the old and new model against them side by side.<\/p>\n On July 23 that middle tier was a judgment call. As of this morning it is the obvious default, and the model id to write into your config is This post was drafted with Empower Network’s AI content engine and edited by a human before publishing.<\/em><\/p>\n Claude Opus 5 launched July 24, 2026 at Opus 4.8 pricing with near-Fable-5 performance. What changed, what it costs, and how operators should use it.<\/p>\n","protected":false},"author":0,"featured_media":59,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[5],"tags":[2,8,6,35],"class_list":["post-58","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai","tag-ai-agents","tag-ai-models","tag-anthropic","tag-claude-opus-5"],"_links":{"self":[{"href":"https:\/\/empowernetwork.com\/ai\/wp-json\/wp\/v2\/posts\/58","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/empowernetwork.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/empowernetwork.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/empowernetwork.com\/ai\/wp-json\/wp\/v2\/comments?post=58"}],"version-history":[{"count":1,"href":"https:\/\/empowernetwork.com\/ai\/wp-json\/wp\/v2\/posts\/58\/revisions"}],"predecessor-version":[{"id":61,"href":"https:\/\/empowernetwork.com\/ai\/wp-json\/wp\/v2\/posts\/58\/revisions\/61"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/empowernetwork.com\/ai\/wp-json\/wp\/v2\/media\/59"}],"wp:attachment":[{"href":"https:\/\/empowernetwork.com\/ai\/wp-json\/wp\/v2\/media?parent=58"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/empowernetwork.com\/ai\/wp-json\/wp\/v2\/categories?post=58"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/empowernetwork.com\/ai\/wp-json\/wp\/v2\/tags?post=58"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}claude-opus-5<\/code>, plus Amazon Bedrock, Google Cloud, and Microsoft Foundry. 9to5Mac reports<\/a> it is the new default on Claude Max plans and the strongest model available to Claude Pro subscribers, so most paying Claude users get it without touching a setting.<\/p>\n![\"Claude](\"https:\/\/empowernetwork.com\/wp-content\/uploads\/sites\/904\/2026\/07\/claude-opus-5-explained-infographic.png\")
<\/figure>\nClaude Opus 5 pricing: same sticker, more work per dollar<\/h2>\n
\n
What changed since Opus 4.8<\/h2>\n
\n
What this means for real work<\/h2>\n
Honest limits, and who should skip it<\/h2>\n
Where Opus 5 fits in the stack<\/h2>\n
claude-opus-5<\/code>.<\/p>\nSources<\/h3>\n
\n