Anthropic just dropped its latest frontier model
Anthropic officially launched Claude Opus 5, the newest addition to its Opus line, aimed squarely at coding, research, and complex business workflows. The company is making a bold claim: it rivals the much-hyped Claude Fable 5 on several key tests — while costing significantly less per task.
Opus 5 is rolling out across Claude’s apps and API. It’s now the default model for Claude Max subscribers and the strongest option bundled with Claude Pro. For users who’ve been watching Fable 5 move behind pay-as-you-go credits starting July 20, this could be a welcome shift.
How close is Claude Opus 5 to Fable 5, really?
Anthropic’s own numbers paint an interesting picture. On CursorBench 3.2 — a benchmark that tests coding agents on real software development tasks — Opus 5 finished within 0.5% of Fable 5’s best score at maximum effort. The catch? It hit that result at half the cost per task.
On OSWorld 2.0, which measures how well AI agents operate computers and complete tasks across apps and files, Opus 5 actually beat Fable 5’s best result at just over a third of the cost. That’s not a small margin.
Still, these are Anthropic’s own tests. Real-world performance can vary wildly depending on your workload, your codebase, and how you prompt the model. Treat the numbers as directional, not gospel.
What the benchmarks actually measure
- CursorBench 3.2: Real software engineering tasks, scored on completion and correctness.
- OSWorld 2.0: AI agents navigating a computer, using apps, and managing files like a human would.
- Frontier-Bench v0.1: Difficult software engineering problems, testing persistence and debugging ability.
Big leap over Opus 4.8
Compared to its predecessor, Opus 5 is a substantial upgrade. Anthropic says it more than doubled Opus 4.8’s score on Frontier-Bench, which tests AI agents on challenging software engineering tasks. It also completed those tasks at a lower average cost.
The company highlights that Opus 5 is better at checking its own work, diagnosing root causes of bugs, and pushing through difficult jobs instead of stopping after a quick fix. In one example, it caught an edge case that an existing community patch had missed. That kind of thoroughness matters when you’re debugging a production system at 2 a.m.
Pricing stays the same
Here’s the part that might surprise you: Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens — unchanged from Opus 4.8. You’re getting a more capable model for the same price. For developers and teams watching API bills, that’s a meaningful win.
For Pro and standard Team subscribers, Opus 5 could fill the gap left by Fable 5 moving behind pay-as-you-go credits. If you’ve been relying on Fable for heavy lifting but don’t want to pay per task, Opus 5 is now your best in-subscription option.
Should you switch to Claude Opus 5?
If you’re a Claude Max or Pro user, you don’t need to do anything — Opus 5 is already there. For API users, it’s worth running your own evals on your specific tasks. The benchmark results are promising, but nothing beats testing on your own code.
Anthropic’s positioning is clear: Opus 5 is the cost-effective workhorse, while Fable 5 remains the premium option for the absolute hardest problems. If your budget is tight and your tasks are complex, Opus 5 deserves a serious look.
Just remember — these are vendor-reported numbers. Independent verification will come as developers push the model in the wild. Until then, consider this a strong contender, not a certified champion.
