News
Guide
Is AI Making Your Organization More Productive or More Dependent?
Key Takeaways
- Claude Opus 5 matches or nearly matches Fable 5 on major coding and knowledge benchmarks at half the per-task cost, according to Anthropic's published results.
- Anthropic calls it their most aligned model yet, with the lowest rates of deceptive behavior in internal audits.
- Opus 5 pricing matches Opus 4.8 ($5/$25 per million tokens), and is available now on all platforms.
Anthropic has released Claude Opus 5, the new cost-efficient alternative to Claude Fable 5. Anthropic pitches Opus 5 as an everyday workhorse, approaching Fable 5's benchmark performance at roughly half the cost per task, according to Anthropic.
Opus 5 is now the default model on Claude Max and the "strongest model" available on Claude Pro, according to Anthropic. It's priced at $5 per million input tokens and $25 per million output tokens, unchanged from Opus 4.8.
Opus 5 Benchmarks Show a Big Leap Over Opus 4.8
"Claude Opus 5 is a clear generational step up from Opus 4.8. Over one weekend I gave it a chief-of-staff role over my dev environments: it built its own monitor, drove each box, and pulled me in only for the judgment calls."
- Cristian Rivera
Staff Software Engineer, Stripe
Across coding, problem-solving and knowledge work, Anthropic's published benchmarks show Opus 5 outscoring its predecessor and closing the gap with Fable 5.
"Claude Opus 5 delivers the industry intelligence and accuracy that is essential for the analysis of specialized enterprise content," said Ben Kus, CTO at Box. At his company, said Kus, Opus 5 outperformed 4.8 by 8% and delivered performance gains in data analysis (11% improvement) and due diligence (17% improvement) workflows that technology, healthcare and public sector organizations rely on.
Coding benchmarks:
"Claude Opus 5 delivers near Fable 5 intelligence at Opus speed and cost," said Sualeh Asif, co-founder of Cursor. "On CursorBench, it's just under Fable 5 and has many of the same behaviors."
- On Frontier-Bench v0.1, Opus 5 is the top model, more than doubling Opus 4.8's results at a lower cost per task.
- On CursorBench 3.2 at max effort, it lands within 0.5% of Fable 5's peak score, at half the cost.
Knowledge and problem-solving:
| Benchmark | Opus 5 Result |
|---|---|
| ARC-AGI 3 (novel problem solving) | 3× the score of the next-best model |
| Zapier AutomationBench (business task completion) | ~1.5× the next-best model's pass rate at equal cost |
| OSWorld 2.0 (computer use) | Beats Fable 5's best result at roughly one-third the cost |
Opus 5 also shows notable improvements in scientific research, per Anthropic's internal benchmarks. On organic chemistry tasks such as inferring molecular structures from spectroscopy data, Opus 5 scores 10.2 percentage points above Opus 4.8. Protein-related tasks saw a 7.7-point improvement.
What Working With Opus 5 Looks Like
Claude Opus 5 can verify its own work and iterate until a task actually succeeds, Anthropic noted. Several examples from evaluations and early-access testing:
- Given a drawing of a machine part but no direct way to view it, Opus 5 built its own computer vision pipeline to extract the geometry from raw pixels and then reconstructed the full 3D model in FreeCAD. No competing model could solve the same task.
- On a real bug in a popular open-source package manager, Opus 5 identified and fixed the root cause — including an edge case that a community patch had missed — while a competitor only addressed the surface symptom.
- An engineer at a trading firm used Opus 5 to build a complete market data feed for a new exchange in a single session, a task previous models couldn't complete even with detailed plans provided upfront.
How Opus 5 Handles Safety and Misuse
Anthropic's internal behavioral audit ranks Opus 5 as what the company calls its "most aligned model to date." It adheres more closely to Claude's Constitution than Opus 4.8, Sonnet 5 or Fable 5, and shows the lowest rates of deceptive behavior.
On the safety front, the company has deliberately kept Opus 5 behind Mythos 5 in offensive cybersecurity. While Opus 5 can identify software vulnerabilities at a similar rate to Mythos 5, it is significantly less capable at developing working exploits.
Key safeguard details:
- Opus 5's cyber classifiers are less restrictive than Fable 5's, intervening roughly 85% less often, but still block penetration testing and exploit generation.
- Flagged requests in Claude.ai and Claude Code fall back to Opus 4.8 by default.
- Enterprises enrolled in Anthropic's Cyber Verification Program get access to a version with fewer restrictions.
What Else Ships Alongside Opus 5
Two beta features launch with the model:
- Mid-conversation tool changes on the Claude Platform, letting developers swap available tools without breaking the prompt cache.
- Automatic API fallbacks, where requests flagged by safety classifiers route to the next best available model instead of being blocked outright.
A Fast mode is also available, running approximately 2.5× the default speed at double the base price.
Editor's Note: For more on the evolving tradeoffs between AI model cost and performance...
- Cursor Launches AI Model Router That Cuts Developer Costs by Up to 50% — Cursor launches intelligent model routing system that automatically selects the best AI model per task.
- AI Model Prices Are Falling At The Worst Moment For The US Frontier Labs — The price war is on. What happens to OpenAI and Anthropic?
- Kimi K3's 2.8T Open Model Exposes the Fragile Economics Behind AI's Business Model — If a free model can match Claude and GPT, who pays for the $250 billion in data centers being built to run them?