Claude Opus 5 is Anthropic’s newest AI model, released on July 24, 2026, and available across all of the company’s platforms. It’s positioned as a “daily driver” model that comes close to the intelligence of Anthropic’s most powerful generally available model, Claude Fable 5, while costing roughly half as much per task. According to Anthropic’s official announcement, Opus 5 sets new state-of-the-art results on coding and knowledge work evaluations, becomes the default model on Claude Max, and is now the strongest model available on Claude Pro.
In this article we’ll discuss what Claude Opus 5 is, how it performs on coding, knowledge work, and scientific research benchmarks, and why its pricing and adjustable “effort” setting matter so much to businesses worried about runaway AI bills. We’ll also look at what early customers are reporting, how Opus 5 fits into Anthropic’s rapid-fire release schedule, and what the company says about the model’s alignment and safety profile.
TL;DR Snapshot
Claude Opus 5 represents a shift in how AI labs are competing. Rather than chasing raw capability alone, Anthropic is betting that the most valuable AI work happens in a middle band of difficulty where near-frontier intelligence, delivered cheaply and reliably, wins the day. The model keeps the same pricing as its predecessor, Opus 4.8, while more than doubling performance on some benchmarks, and it ships with features designed to give customers control over how much they spend on each task.
Key takeaways include…
- Opus 5 delivers performance close to Anthropic’s flagship Fable 5 at roughly half the cost, with pricing unchanged from Opus 4.8 at $5 per million input tokens and $25 per million output tokens.
- A new “effort” dial lets users choose how much computing power the model spends on a task, directly addressing enterprise concerns about expensive AI bills.
- Anthropic says Opus 5 is its most aligned model to date, scoring lowest on its internal misaligned-behavior audit while deliberately staying behind Mythos 5 on risky dual-use capabilities like offensive cybersecurity.
Who should read this: Developers, enterprise IT leaders, startup founders, and AI enthusiasts tracking the economics of frontier models.
A Model Built for Every Day, Not Just the Hardest Problems
The most interesting thing about Claude Opus 5 is it’s positioning. As VentureBeat reported, Anthropic isn’t claiming that Opus 5 is its smartest model. That title still belongs to Claude Fable 5/Mythos 5. Instead, the company is arguing that the most economically important AI work sits in a middle tier of difficulty, where efficiency matters as much as intelligence, and this launch signals how the AI race is shifting away from raw capability and toward the economics of daily use.
That framing shows up in the product decisions as well. Anthropic’s announcement describes Opus 5 as a model designed to be used every day, working more efficiently than other models. It’s now the default selection on Claude Max, Anthropic’s premium consumer tier, and the top option on Claude Pro. Developers can access it through the Claude API under the model string claude-opus-5, and there’s a Fast mode that runs at roughly 2.5 times the default speed for twice the base price.
The context helps explain the strategy. According to Fortune, Opus 5 is the fourth model Anthropic has shipped in just under two months, following Mythos 5, Fable 5, and Sonnet 5 in June. Fortune also notes that business customers had criticized Fable 5’s high token burn rate, which caused some users to blow through their token budgets or rack up large bills. Opus 5, with its efficiency focus and unchanged pricing, reads as a direct answer to that complaint.
Benchmarks, Effort Dials, and the New Economics of AI
On paper, the performance gains are substantial. Per Anthropic, Opus 5 sets new state-of-the-art results among its models on coding and knowledge work evaluations like Frontier-Bench and GDPval-AA. On Frontier-Bench v0.1, it more than doubles Opus 4.8’s performance at a lower cost per task. On ARC-AGI 3, a test of novel problem solving, its score is three times that of the next-best model. On OSWorld 2.0, a computer use benchmark, it beats Fable 5’s best result at just over a third of the cost. And on Zapier’s AutomationBench, which measures whether models can complete business tasks end to end, its pass rate is around 1.5 times the next-best model at the same cost per task.
The headline feature for cost-conscious buyers, though, is the effort setting. As Fortune reported, users can toggle between low, medium, and high effort levels to balance cost against capability. Axios adds that at lower effort levels, Opus 5 can preserve much of its performance while using fewer tokens and costing less to run. In other words, customers get a dial for the tradeoff that used to be baked into which model they picked.
Early customers have been candid about their experiences. In Anthropic’s launch post, legal AI company Harvey’s head of applied research, Niko Grupen, reported that Opus 5 matched high-end performance while generating 26% fewer tokens on average than Opus 4.8 at max reasoning. Box CTO Ben Kus said that Opus 5 outperforms Opus 4.8 by 8% overall, with an 11% improvement on data analysis workflows and 17% on due diligence. Zapier CEO Wade Foster reported that Opus 5 topped the company’s AutomationBench leaderboard and passed a full churn-prevention workflow that previous models had failed. And Cognition CEO Scott Wu noted that within their Devin solution, the model showed particular strength on difficult debugging and root-cause analysis.
The model’s thoroughness also came through in Anthropic’s own testing. In one example from their official announcement, Opus 5 was asked to rebuild a machine part as a 3D model from a drawing it had no way to directly view, so it wrote its own computer vision pipeline to extract the geometry from raw pixels and reconstructed the part, something no competing model managed in five attempts under the same setup.
Alignment, Safety, and What Anthropic Deliberately Left Out
Anthropic paired the capability story with an alignment one. According to their launch announcement, an automated behavioral audit found Opus 5 to be their most aligned model to date, scoring 2.3 on overall misaligned behavior, the lowest of its recent models. The company says the model adheres to Claude’s Constitution better than Opus 4.8, Sonnet 5, or Fable 5, shows the lowest rates of deceptive behavior, and is the least susceptible to being tricked into misuse.
Just as notable is what Opus 5 doesn’t do. Anthropic says the model doesn’t advance the frontier in risky, dual-use capabilities, remaining behind Mythos 5 (the restricted-access counterpart to Fable 5) in both biology research and offensive cybersecurity. The company intentionally avoided training Opus 5 on cyber tasks. On its OSS-Fuzz evaluation, Opus 5 identifies software vulnerabilities about as well as Mythos 5 but falls far behind at actually developing exploits for them.
The safeguards reflect that profile. Per Anthropic, Opus 5’s cyber classifiers are proportionally less restrictive than Fable 5’s, allowing vulnerability discovery in source code while blocking binary-based scanning, penetration testing, and exploit generation. The company expects these classifiers to intervene around 85% less often than they do for Fable 5. Alongside the launch, Anthropic also introduced automatic fallbacks in beta, letting API requests flagged by safety classifiers route to another available model instead of being blocked.
The safety conversation around this release comes on the heels of a tumultuous Fable/Mythos launch. As Fortune reported, Fable 5 was temporarily subject to U.S. government export controls after Amazon researchers found they could bypass its safeguards, prompting Anthropic to pull the model on June 12 and re-release it on June 30 with strengthened security measures. And Axios noted that Anthropic is working with government partners to conduct independent testing of its models, including Opus 5. Against that backdrop, shipping a highly capable model that deliberately stays inside known safety boundaries looks like a strategic choice, not just a technical one.
Frequently Asked Questions
Anthropic is an AI company founded in 2021 by former OpenAI staff. It’s the maker of the Claude family of AI models, which includes Mythos, Fable, Opus, Sonnet, and Haiku tiers, and it operates products like Claude.ai, Claude Code, and Claude Cowork.
Claude Opus 5 is an AI model released by Anthropic in July 2026. It’s designed for coding, knowledge work, agentic tasks, and scientific research, and it delivers performance close to Anthropic’s flagship Fable 5 model at about half the cost per task. It’s available on all of Anthropic’s platforms, including Claude Max, Claude Pro, and the Claude API.
Claude Fable 5 is Anthropic’s most powerful generally available model, released in June 2026. It remains the company’s recommendation for the most advanced projects, but it drew criticism from business customers for its high token consumption, and it was briefly pulled from the market over security concerns before being re-released with strengthened safeguards in place.
Claude Mythos 5 is a version of Claude Fable 5, Anthropic’s most capable model, which is only available to approved organizations. It shares the same underlying model as Fable 5 but without some of the additional safety measures, and Anthropic uses it as the reference point for dual-use capabilities like biology research and offensive cybersecurity.
Opus 5 is priced at $5 per million input tokens and $25 per million output tokens, the same as its predecessor, Opus 4.8. A Fast mode running at roughly 2.5 times the default speed is available at twice the base price.
Tokens are the units of information an AI model processes. In English text, a token works out to roughly one short word or a fraction of a long word. They’re the unit AI companies typically use to bill for model usage.
The effort setting is a dial that lets users control how much computing power Opus 5 devotes to a task. Higher effort means more intelligence applied to harder problems, while lower effort conserves tokens for faster, cheaper results with slightly reduced performance.
Other Enterprise AI Articles You May Be Interested In
OpenAI AI Models Hacked Hugging Face: Full Breakdown
Meta AI Layoffs Lawsuit: 26 Employees Allege Discrimination
Nvidia Bets Big on Japan: Inside the New Robotics and Physical AI Partnerships
Microsoft Becomes First Hyperscaler to Adopt 3M EBO Technology
