Claude Opus 5 Released: Near-Fable Intelligence at Half the Price

What Claude Opus 5 brings to the table

Anthropic released Claude Opus 5 on July 24, 2026, positioning it as the new default model on Claude Max and the strongest model available on Claude Pro. The headline is straightforward: it delivers near-Fable 5 intelligence at roughly half the cost per task, while keeping the exact same token pricing as its predecessor Opus 4.8.

The model is built for everyday use. Anthropic describes it as "thoughtful and proactive" — it verifies its own work, catches errors before they compound, and iterates through complex tasks without needing a human to hold its hand. Early-access customers across finance, legal, biotech, and software engineering fields report that Opus 5 handles tasks previous models simply could not complete.

On Frontier-Bench v0.1, Opus 5 more than doubles Opus 4.8's performance while costing less per task. On CursorBench 3.2, it lands within half a percentage point of Fable 5's peak score at maximum effort — but at half the price, according to Anthropic's official announcement. Cursor co-founder Sualeh Asif confirmed the model "delivers near Fable 5 intelligence at Opus speed and cost" and "has many of the same behaviors."

Benchmarks: where Opus 5 stands

Independent measurements from Artificial Analysis place Opus 5 at the very top of the Intelligence Index v4.1 with a score of 61 — one point ahead of Fable 5 (60) and two points ahead of GPT-5.6 Sol at maximum reasoning (59). Here is how the current frontier stacks up:

Data: Artificial Analysis Intelligence Index v4.1 (July 24, 2026). EUR conversions at ~0.92 USD/EUR, rounded. Output prices: Opus 5 = $25/€23.00, Fable 5 = $50/€46.00, GPT-5.6 Sol = $25/€23.00.

The standout individual result is ARC-AGI 3, where Opus 5 scored three times higher than the next-best model — a gap that signals genuinely new reasoning capability, not just incremental improvement. On OSWorld 2.0 (a computer-use benchmark), Opus 5 surpassed Fable 5's best result while consuming roughly one-third the compute budget.

Perhaps the most revealing anecdote from the launch: on one Frontier-Bench task, Opus 5 was given a drawing of a machine part but no way to view it. The model wrote its own computer-vision pipeline to extract geometry from raw pixels, then reconstructed the part in FreeCAD — and did so repeatedly. No competing model solved this task in five attempts.

The European picture: availability, pricing, and the competition gap

Claude Opus 5 is available today in all 27 EU member states. Anthropic serves European customers through Anthropic Ireland Limited, the company's EU legal entity. The model is accessible via claude.ai, the Claude API, and cloud partners — including AWS (with EU region deployments available), Google Cloud Vertex AI, and Microsoft Foundry.

For European developers and enterprises, the key numbers are these: at $5 / €4.60 per million input tokens and $25 / €23.00 per million output tokens, Opus 5 costs the same as Opus 4.8 while delivering dramatically higher performance. A typical 10 000-token coding task with Opus 5 costs roughly €0.25 in output — half what the same-quality result would cost with Fable 5 (€0.50). Cached inputs drop the cost by 90% to $0.50 / €0.46 per million tokens, making repeated queries within the same 1-million-token context window extremely economical.

One uncomfortable fact emerges clearly from the data: no European AI company operates at this performance level. Mistral, the leading EU model maker, does not appear in the top tier of the Artificial Analysis Intelligence Index. Aleph Alpha has pivoted away from frontier models toward enterprise tooling. The closest thing Europe has to a frontier model provider is Anthropic Ireland Limited itself — and even that is a subsidiary of a US company.

This matters in practice. The EU AI Act takes full enforcement effect on August 2, 2026, giving the EU AI Office real powers over general-purpose AI models — including fines of up to 3% of global turnover for non-compliance (we covered the details in our AI Act enforcement guide). European businesses that rely on frontier models like Opus 5 need to understand what compliance obligations they carry as deployers, and what Anthropic handles on its side as the provider.

Safety: deliberately limited on cyber exploitation

Anthropic's safety approach with Opus 5 is notable. The model was intentionally not trained on cybersecurity exploitation tasks — a conscious decision to keep it behind Mythos 5 in offensive cyber capabilities. On the OSS-Fuzz evaluation, Opus 5 is nearly as good as Mythos 5 at finding software vulnerabilities, but substantially worse at exploiting them. As Anthropic puts it in the Opus 5 System Card, the model improved on cyber tasks only "as a result of becoming more generally capable," not through targeted training.

The model also scores as Anthropic's most aligned to date: it achieved the lowest misaligned-behavior score (2.3) of any recent model on the company's automated behavioral audit, exhibited the lowest rates of deceptive behavior, and proved the most resistant to being tricked into misuse.

For Claude.ai, Claude Code, and Claude Cowork users, flagged security-related requests automatically fall back to Opus 4.8 rather than being blocked outright — a practical compromise that keeps developers productive while maintaining guardrails. Companies in Anthropic's Cyber Verification Program get access to a less-restricted version of Opus 5 immediately.

What developers and early users are saying

The early-access testimonials from Anthropic's launch paint a consistent picture: Opus 5's "judgment" is what sets it apart. A principal AI engineer at a tech firm reported that the model pushed back on a flawed design proposal, narrowed its objection to a single design question, and proposed a compromise — "that's the kind of judgment that lets us trust it with less oversight." A staff engineer at another company gave Opus 5 a "chief-of-staff role" over his development environments over a single weekend: it built its own monitor, drove each box, and pulled him in only for judgment calls.

On the quantitative side, Box reported an 8% overall improvement over Opus 4.8, with 11% gains in data analysis and 17% in due diligence workflows. A trading firm saw 9 percentage points higher accuracy with a third fewer tool calls and 60% less time on complex financial-modeling tasks.

We covered Anthropic's recent voice mode update and app connectors in our earlier article today — those features remain available with Opus 5, and the model upgrade makes the entire Claude ecosystem significantly more capable for European productivity users.

The bottom line

Claude Opus 5 resets expectations for what a "practical" frontier model should deliver. It matches or exceeds much more expensive models on real-world coding and knowledge-work tasks while costing half as much as Fable 5. For Europeans, the model is fully available and backed by Anthropic's EU legal entity — but the broader strategic picture is that no EU-headquartered company can offer anything close to this capability level. With the AI Act's enforcement deadline just over a week away, that gap is more than an academic concern.