Claude Opus 4.8 is here. Is it as good as they say?

Key takeaways

Opus 4.8 performs very well on initial, surface-level tasks like prototypes and simple code, but struggles with the last 10%, particularly Edge cases and bug fixing.
The model tends to hallucinate and over-rely on hypotheses without grounding in factual data, especially in complex or edge scenarios.
It's fast, efficient, and offers a great user experience, making it ideal for quick prototyping, initial design, and tool integration.
For data-heavy, strategic, or deeply contextual work, Opus 4.7 currently remains more reliable due to its better anchoring in real data.
New features like dynamic workflows and effort control are significant additions, though their full effectiveness will need further testing.

Summary

Introduction to Claude Opus 4.8 and Early Access Insights

Clarevo introduces the episode by sharing her early access experience with Anthropic's latest model, Opus 4.8. She discusses the context of its release, the enthusiasm for testing new features, and sets the stage for a comprehensive review. The section highlights her focus on real-world coding, strategic, and design tasks to evaluate the model's performance.

Performance and Benchmarks of Opus 4.8

This section covers the official benchmarks and performance metrics of Opus 4.8, noting its improvements over previous versions and competitors. The model achieves a significantly higher success rate on benchmarks (around 69.2%) and is positioned as a more honest, instruction-following, enterprise-ready model. Pricing is also discussed, emphasizing its cost at $5 per input token and $25 per million output tokens.

Coding Capabilities and Limitations

Clarevo tests Opus 4.8's coding skills through various projects: building a prototyping tool, rebase tasks with existing codebases, and creating engaging games for a young audience. She notes the model's strengths in initial implementation but points out its struggles with the last 10%—bugs, hallucinations, and difficulty handling edge cases—highlighting a tendency for overconfidence and hallucination based on hypothesis rather than verified data.

Business Strategy and Analytical Performance

The model's performance in strategic tasks is evaluated by comparing Opus 4.7 and 4.8. Clarevo finds that Opus 4.8 tends to be more hand-wavy and less anchored in data, often hallucinating or over-rotating on small data points. This inconsistency affects its reliability in strategic thinking and roadmapping, making her prefer the more grounded Opus 4.7 for data-heavy and analytical tasks.

Positives and Final Verdict

Despite its limitations, Opus 4.8 impresses with its voice quality, fast response times, and user experience, which is smooth and non-annoying. Clarevo appreciates its efficiency and easy-to-read output. She concludes that while the model is not revolutionary, it’s a solid, fast, and tool-oriented option suitable for prototypes, tool use, and initial feature development. Ongoing testing is encouraged to better understand its capabilities and quirks.

Show notes

I got a few hours of early-access testing with Anthropic’s newly released model Opus 4.8. I walk through real coding, design, and strategy tasks across Claude Code and Claude Cowork, and give you my unfiltered view on what impressed me and what didn’t.

—

What you’ll learn:

Where Opus 4.8 excels: greenfield prototypes, one-shot features, and fast execution
Where it struggles: the last 10%, edge cases in existing codebases, and hallucinations
How Opus 4.8 compares to Opus 4.7 on business strategy work
Why I’m still reaching for Opus 4.7 on data-heavy strategy and roadmap work
The new features shipping alongside the model: dynamic workflows with parallel subagents and effort control in Claude.ai and Cowork
The prompting and harness strategy I’d use to get the most out of it

—

In this episode, we cover:

(00:00) Introduction to Opus 4.8

(00:44) Benchmark performance and pricing

(01:53) First coding test: Building a prototyping tool

(03:00) Where it failed: The last 10% problem

(03:27) The hallucination problem

(04:23) Testing Opus 4.8 on existing codebases

(05:24) The ambition test: Building games for a 9-year-old

(07:03) Business strategy test: 4.7 vs 4.8

(08:23) The roadmap test

(09:17) Final verdict