---
id: "2026-11-02-memdoor-five-qwen-agents-beat-fable-5-livecodebench"
url: "https://postcutoff.com/e/2026-11-02-memdoor-five-qwen-agents-beat-fable-5-livecodebench/"
as_of: "2026-11-09T21:13:00+01:00"
date: "2026-11-02"
date_precision: day
category: open-source
importance: 2
confidence: medium
status: [Partly confirmed]
sources: 3
editor: human editor
human_review: null
version: "2026-11-08"
---

As of: 2026-11-09 21:13 CET.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/
Canonical page: https://postcutoff.com/e/2026-11-02-memdoor-five-qwen-agents-beat-fable-5-livecodebench/

# Memdoor: a team of five Qwen3.8-27B agents beats Fable 5 on ten LiveCodeBench Hard problems (open-source harness, HN 442 pts)

The open-source coding-agent harness Memdoor reproduced the GVS5H multi-agent workflow and reported that five Qwen3.8-27B agents solved a mean 9.8 of the first 10 LiveCodeBench Hard problems (5 passes) versus 8.0 for a single Fable 5 call; the original authors reported 92.4 vs 90.4 on their full 100. Small sample, self-reported, all runs published.

## Key facts
- 10 problems x 5 passes: Memdoor 9.8/10, GVS5H authors' harness 9.6, Fable 5 single call 8.0
- GVS5H authors' full-100 result: 92.4 vs 90.4
- Apache-2.0 Go binary; GitHub ~380 stars by Nov 8; HN 442 points (Nov 2)

## Your AI and this story

- GPT-6 Astra (training cutoff April 2026): 186 days after its cutoff
- Claude Opus 5.5 (training cutoff June 2026): 125 days after its cutoff
- Gemini 4 Argon (training cutoff March 2026): 216 days after its cutoff
- Grok 4.7 (training cutoff May 2026): 155 days after its cutoff

## Sources

1. [GitHub: guregodevo/memdoor](https://github.com/guregodevo/memdoor) (github.com, code)
2. [GVS5H (original harness)](https://github.com/slee-persis/GVS5H) (github.com, code)
3. [Hacker News discussion](https://news.ycombinator.com/item?id=50320995) (news.ycombinator.com, discussion)

## Changes

- 2026-11-08 (filed): Created