Claude Fable 5 narrowly beats Opus 4.8 in coding-agent test

A Reddit user says they compared Claude Fable 5 and Opus 4.8 across 917 coding-agent scenarios. The post says Fable 5 scored 92.9 overall, while Opus 4.8 scored 92.0. It also says Fable 5 cost about $1.25 per task, while Opus 4.8 cost about $0.74 per task. The user says Fable 5 refused 26 tasks that Opus completed successfully.

Key points

  • The test used 917 coding-agent scenarios, according to the post.
  • Fable 5 scored 92.9 overall, while Opus 4.8 scored 92.0.
  • The reported cost was about $1.25 per task for Fable 5 and $0.74 for Opus 4.8.
  • The post says Fable 5 refused 26 tasks that Opus completed.
  • The main question raised is whether the small score gain is worth the higher cost.

Quick term guide

Claude Fable 5
The name of an AI tool or model mentioned in the post, but the item does not give enough information to verify details.
Claude Fable
A new Claude AI model released by Anthropic in June 2026
Opus 4.8
A high-end Claude AI model named by the project.
coding-agent scenarios
Test tasks where an AI tool is asked to handle coding work like an assistant.
coding-agent
An AI tool that can help write, edit, and reason about code across a task.
Solo developer
An individual who handles all parts of creating a project or product alone.
solo dev
A person who builds software mostly by themselves.
workflow
A repeatable set of steps for getting a task done.
Read original