Table of Contents
Anthropic has a new answer to a question a lot of AI teams have been quietly asking all year: what do you use for the 95% of work that doesn’t need your most expensive model? On July 24, 2026, the company released Claude Opus 5, positioning it not as a new frontier flagship, but as something more practical — a model designed to be the one you reach for every single day.
According to Anthropic’s official announcement, Opus 5 is a “thoughtful and proactive model that comes close to the frontier intelligence of Claude Fable 5 at half the price.” That single sentence captures the whole strategy: rather than chasing another headline-grabbing capability jump, Anthropic built a model meant to make near-frontier intelligence affordable enough to run constantly, across real workdays, without triggering a budget review.
This is Anthropic’s fourth model release in under two months — following Mythos 5, Fable 5, and Sonnet 5, all shipped in June 2026 — and it lands at a moment when enterprise AI spending is under real scrutiny. Businesses have grown wary of unpredictable token bills, and Opus 5 is Anthropic’s direct response to that pressure.
What Is Claude Opus 5?
Claude Opus 5 is the newest model in the Opus tier, sitting below Anthropic’s most capable model, Fable 5, and its restricted-access model, Mythos 5, in the company’s four-tier Claude 5 lineup: Haiku, Sonnet, Opus, and Fable, with Mythos occupying a separate, tightly access-controlled category above them.
Rather than trying to out-benchmark Fable 5 outright, Opus 5 is built around a different goal: getting as close as possible to Fable-level performance while running at roughly half the cost and with meaningfully better efficiency. Anthropic describes it as designed for daily use — handling long-running agents, coding tasks, and professional knowledge work reliably enough that teams don’t need to reach for a pricier model out of habit.
It’s now the default model on Claude Max and the strongest model available on Claude Pro, which means most everyday Claude.ai users are already interacting with it, often without needing to select it manually.
Key Features and Capabilities
Effort Settings: Trading Intelligence for Cost
One of the most practical additions in Opus 5 is an adjustable effort setting, which lets developers and users choose how much reasoning the model applies to a given task. Anthropic’s benchmark charts show performance scaling directly with effort level, so a team can dial the model down for routine tasks and up for harder ones — without switching models entirely.
This matters because it turns “which model should I use” into “how much effort does this task deserve,” a much more granular and cost-efficient way to manage AI spend.
Stronger Agentic Judgment
Anthropic emphasizes that Opus 5 is notably better at verifying its own work and recovering from errors mid-task, rather than needing a human to catch mistakes. The official announcement includes several concrete examples:
- Given a machine-part drawing with no direct way to view it, Opus 5 wrote its own computer vision pipeline to extract the geometry from raw pixels and rebuilt the part as a 3D model — succeeding on repeated attempts where competing models failed even after five tries.
- When fixing a real bug in a popular open-source package manager, Opus 5 identified and corrected the actual root cause, while a competing model patched only the surface symptom.
- An engineer at a trading firm reportedly used Opus 5 to build a market data feed for a new exchange in a single session — a task earlier models couldn’t complete even with detailed guidance. Opus 5 went a step further and built its own test harness to validate the code against the exchange’s data format.
Life Sciences and Research Gains
Opus 5 also shows meaningful improvement over Opus 4.8 across Anthropic’s internal life sciences evaluations, spanning structural biology, organic chemistry, and bioinformatics. The company reports its biggest gains on organic chemistry tasks like inferring molecular structure from spectroscopy data, and on predicting how sequence variations affect protein function.
Visual and Front-End Output
Anthropic also highlights stronger visual generation, showcasing interactive demos Opus 5 built independently — including a physics-based wind tunnel simulation and an interactive illustrated cell model — both cited as evidence of improved front-end and visual reasoning over prior Opus versions.
Developer-Facing Betas
Alongside the model, Anthropic shipped two beta features on the Claude Platform:
- Mid-conversation tool changes — developers can now change which tools Claude has access to mid-conversation without invalidating the prompt cache, reducing latency and cost overhead in longer agent sessions.
- Automatic fallbacks — API requests flagged by safety classifiers can now automatically route to another available model instead of being blocked outright, so workflows keep running rather than stalling on a refusal.
Claude Opus 5 vs. Claude Opus 4.8 vs. Claude Fable 5
| Feature | Opus 4.8 | Opus 5 | Fable 5 |
|---|---|---|---|
| Role in lineup | Previous daily-use model | New default daily-use model | Most capable public model |
| Price (per 1M tokens) | $5 in / $25 out | $5 in / $25 out (unchanged) | Higher (frontier tier) |
| Frontier-Bench v0.1 | Baseline | More than double Opus 4.8’s score | Reference point Opus 5 approaches |
| CursorBench 3.2 (max effort) | Lower | Within 0.5% of Fable 5’s peak score | Highest overall |
| ARC-AGI 3 | Lower | ~3x the next-best model’s score | N/A (Opus 5 leads this eval) |
| Cybersecurity capability | Limited | Improved, but intentionally not optimized for cyber tasks | Strongest for cybersecurity; still trails Mythos 5 |
| Alignment (internal audit) | Higher misalignment score | Lowest misalignment score of any current Claude model | Not the focus of comparison |
| Fast mode | Available | Available, ~2.5x default speed at 2x base price | Available |
| Best suited for | General daily tasks | Everyday coding, agents, professional/business work | Long-running autonomous, highest-stakes projects |
Benchmark figures reflect Anthropic’s own published results as of the July 24, 2026 announcement.
Pricing: What Does Claude Opus 5 Cost?
Anthropic kept Opus 5’s pricing identical to its predecessor: $5 per million input tokens and $25 per million output tokens on the Claude API. The pitch isn’t a price cut — it’s dramatically more capability at the same price point, which functions as a price cut in practical terms since fewer tokens and fewer retries are typically needed to complete the same task.
Fast mode, which runs the model at roughly 2.5 times its default speed, is available at twice the base price — consistent with how Fast mode was priced under Opus 4.8.
How Opus 5 Handles Safety and Cybersecurity
Anthropic built Opus 5 with safeguards similar to those used on Opus 4.8, with some additional guardrails layered onto a narrow set of cybersecurity tasks.
On the safety side, the company reports that Opus 5 scored as its most aligned model yet on an internal automated behavioral audit — showing the lowest rate of deceptive behavior and the least susceptibility to misuse of any recent Claude model, while also being least likely to take reckless, hard-to-reverse actions.
On cybersecurity specifically, Anthropic says Opus 5 approaches Mythos 5’s ability to identify software vulnerabilities but remains substantially behind Mythos 5 at exploiting them — the distinction the company treats as the more safety-relevant capability. Because of this gap, Opus 5’s cyber-related safety classifiers are less restrictive than Fable 5’s: the model can search for vulnerabilities in source code, but requests involving binary-based scanning, penetration testing, or exploit generation are blocked and, by default, routed to Opus 4.8 instead. Anthropic estimates these classifiers will trigger roughly 85% less often than they do on Fable 5.
Enterprises and researchers enrolled in Anthropic’s Cyber Verification Program get access to a version of Opus 5 with fewer of these restrictions for legitimate security research.
On biology, Opus 5 uses safeguards similar to Opus 4.8’s and is now Anthropic’s most capable generally available model for scientific research — though the company notes it still has real limitations on long, autonomous research tasks, an area where Mythos 5 remains the stronger option.
Pricing: What Does Claude Opus 5 Cost?
Anthropic kept Opus 5’s pricing identical to its predecessor: $5 per million input tokens and $25 per million output tokens on the Claude API. The pitch isn’t a price cut — it’s dramatically more capability at the same price point, which functions as a price cut in practical terms since fewer tokens and fewer retries are typically needed to complete the same task.
Fast mode, which runs the model at roughly 2.5 times its default speed, is available at twice the base price — consistent with how Fast mode was priced under Opus 4.8.
How Opus 5 Handles Safety and Cybersecurity
Anthropic built Opus 5 with safeguards similar to those used on Opus 4.8, with some additional guardrails layered onto a narrow set of cybersecurity tasks.
On the safety side, the company reports that Opus 5 scored as its most aligned model yet on an internal automated behavioral audit — showing the lowest rate of deceptive behavior and the least susceptibility to misuse of any recent Claude model, while also being least likely to take reckless, hard-to-reverse actions.
On cybersecurity specifically, Anthropic says Opus 5 approaches Mythos 5’s ability to identify software vulnerabilities but remains substantially behind Mythos 5 at exploiting them — the distinction the company treats as the more safety-relevant capability. Because of this gap, Opus 5’s cyber-related safety classifiers are less restrictive than Fable 5’s: the model can search for vulnerabilities in source code, but requests involving binary-based scanning, penetration testing, or exploit generation are blocked and, by default, routed to Opus 4.8 instead. Anthropic estimates these classifiers will trigger roughly 85% less often than they do on Fable 5.
Enterprises and researchers enrolled in Anthropic’s Cyber Verification Program get access to a version of Opus 5 with fewer of these restrictions for legitimate security research.
On biology, Opus 5 uses safeguards similar to Opus 4.8’s and is now Anthropic’s most capable generally available model for scientific research — though the company notes it still has real limitations on long, autonomous research tasks, an area where Mythos 5 remains the stronger option.
What Early Users Are Saying
Anthropic published feedback from a range of early-access partners across coding, automation, legal, and finance tools. Several described Opus 5 as a clear step up in reliability and judgment rather than just raw capability — Devin’s team noted particular strength on difficult debugging tasks, Cursor’s team said the model performs just under Fable 5 on their internal benchmark, and Zapier reported Opus 5 topped its AutomationBench leaderboard without using more tokens than earlier Claude models. Multiple partners specifically pointed to Opus 5’s ability to catch its own mistakes mid-task — checking work, questioning its own assumptions, and adjusting course — as the biggest practical difference from Opus 4.8.
Who Should Use Claude Opus 5?
- Development teams running everyday coding and agentic workflows who don’t need Fable 5’s top-end capability for every task.
- Business teams automating multi-step workflows (customer support triage, data analysis, document review) where cost predictability matters.
- Claude Pro and Max subscribers, who now get Opus 5 automatically as their default or strongest available model.
- Enterprises already using Claude in coding tools, IDEs, or agent platforms — many of the partners quoted in Anthropic’s announcement are actively migrating production workloads to Opus 5.
If your work involves days-long autonomous projects, the most advanced coding or cybersecurity tasks, or the outer edge of what Claude can do, Anthropic still recommends Fable 5.
The Bigger Picture: Why This Launch Matters
Opus 5 arrives amid a broader industry pivot toward efficiency. OpenAI made similar arguments about token efficiency when it launched its GPT-5.6 family earlier this month, and Google emphasized reduced token usage in its recent Gemini 3.6 Flash release. Across the industry, the “bigger is better” era of pure benchmark chasing is giving way to a harder question from enterprise buyers: what does this actually cost to run at scale, every day?
Anthropic’s own framing supports that read — the company positions Opus 5 explicitly as a response to enterprise cost pressure, not as a capability record attempt. That’s a notable shift in how frontier labs are choosing to compete.
Frequently Asked Questions
- Is Claude Opus 5 better than Claude Opus 4.8? Yes. Anthropic reports that Opus 5 more than doubles Opus 4.8’s score on Frontier-Bench v0.1, shows major gains across knowledge-work and life-sciences evaluations, and scores as Anthropic’s most aligned model to date — all at the same API price.
- How much does Claude Opus 5 cost? $5 per million input tokens and $25 per million output tokens — identical to Opus 4.8’s pricing. Fast mode costs twice the base price.
- Is Claude Opus 5 the same as Claude Fable 5? No. Fable 5 remains Anthropic’s most capable publicly available model. Opus 5 is designed to come close to Fable 5’s performance on many everyday tasks at roughly half the cost, but Anthropic still recommends Fable 5 for the most demanding, long-running, or highest-stakes projects.
- Can Claude Opus 5 be used for cybersecurity research? To a degree. Opus 5 can identify vulnerabilities in source code, but requests involving binary-based scanning, penetration testing, or exploit generation are blocked by safety classifiers. Researchers enrolled in Anthropic’s Cyber Verification Program can access a less-restricted version for legitimate security work.
- Where can I access Claude Opus 5? It’s available today across Claude.ai (as the default model on Claude Max and top model on Claude Pro), the Claude API as claude-opus-5, and through Anthropic’s cloud partners.
- Does Claude Opus 5 replace Claude Sonnet 5? No — Sonnet 5 and Opus 5 sit at different tiers of Anthropic’s lineup, with Opus 5 aimed at more demanding day-to-day professional and agentic work.
Official Sources
- Anthropic — Introducing Claude Opus 5
- Anthropic — Claude Opus 5 System Card
- Anthropic — Claude Opus 5 Prompting Guide
- Anthropic — Claude Models Overview
- Anthropic — Claude PricingAnthropic Newsroom