Opus 5 spawns 17 subagents and burns 1.5M cached tokens on a simple lookup, Reddit user warns
tr14l · reddit · 2026-09-22
A Reddit user reports that Opus 5 pointlessly shotgun-blasts subagents on simple tasks: a basic lookup of the latest Qwen models and benchmarks spawned 17 agents and triggered 1.5M cache reads, torching the token budget for a three-paragraph answer.
The author says switching to Fable eliminated the problem entirely — it's slightly more expensive per token but usage stays predictable with almost no refresh waits. They also suspect Opus's behavior has degraded since 4.6 ("4.8 was usable, 5 is not trustworthy"), and wonder if runaway background subagents explain the wave of "something is up with Anthropic usage" posts.
TL;DR: use Fable; Opus will blow through your usage for no reason.
More from coding & agent
- Boubou: open-source CLI turns your laptop into a fully local voice AI hub — alexcovo_eth · 2026-11-12
- User automates entire video clipping workflow with Nous Research's Hermes agent — Teknium · 2026-11-12
- As models improve, writing good evals gets harder and more important — amankhan · 2026-11-12
- Prime agent ships with day-one multi-agent communication, runs in Minecraft unmodified — sarahcat21 · 2026-11-12
- Autonomous agent edits just 1,024 weights to lift Qwen3.5 0.8B from 7.4% to 37.4% on MATH-500 — VictorSanh · 2026-11-12
- Omarchy 4.6 lets you share local AI models across your tailnet, says DHH — alexcovo_eth · 2026-11-12