# Opus 5 scores 24% on SlopCodeBench — Humanlayer flags 93% of the code

> An independent third-party benchmark of Claude Opus 5's coding quality lands with a critical read.

- Org: Humanlayer
- Date: 2026-07-27
- Category: article
- Source: https://github.com/humanlayer/advanced-context-engineering-for-coding-agents/blob/main/benchmarking-opus-5-on-slop-code-bench.md
- Canonical page: https://ai-tldr.dev/releases/humanlayer-opus-5-slopcodebench/

## In short

Dex Horthy, founder of Humanlayer, benchmarks Claude Opus 5 on SlopCodeBench and finds a 24% strict pass rate, nearly 5× more code than Opus 4.8, and quality warnings on 93% of the output.

## What it is

Humanlayer's Dex Horthy ran Claude Opus 5 through SlopCodeBench, a benchmark that feeds a coding agent requirements in sequential checkpoints instead of dumping the full spec up front. Opus 5 cleared 4 of the 17 checkpoints across three problems — a 24% strict pass rate.

## How it works

SlopCodeBench measures how well a model absorbs new requirements without dragging past defects forward. Each checkpoint adds a change on top of the previous version of the code, so a small regression in checkpoint 2 fails everything after it. Horthy also tracked complexity metrics, warning counts, and total code volume per problem.

## Why it matters

The Opus 5 launch page led on higher scores at coding, so the SlopCodeBench read is a useful counterweight. Opus 5 wrote almost five times as much code as Opus 4.8 per problem, and 93% of that code triggered quality warnings — with complexity climbing across checkpoints. For developers deciding whether to make Opus 5 their default coding agent, the concrete failure mode is a starting point.

## Who it's for

Developers evaluating Opus 5 for agentic coding, benchmark authors.

## Links

- [The Write-Up](https://github.com/humanlayer/advanced-context-engineering-for-coding-agents/blob/main/benchmarking-opus-5-on-slop-code-bench.md)
- [Author](https://github.com/dexhorthy)
- [Humanlayer](https://www.humanlayer.dev/)
