Two-person lab Altworld open-sources writing model Hemmingway-1

Altworld, a two-person lab based in Switzerland and South Africa, released Hemmingway-1: a writing-only open model fine-tuned from Qwen3.8-27B under Apache-2.0, with weights available on Hugging Face. It scores 1330 on EQ-Bench 4, adding another locally deployable option to the open-source creative writing field.

Confirmed

• Released by Altworld, fine-tuned from Qwen3.8-27B with Apache-2.0 weights on Hugging Face, with sample outputs
• Positioned as a writing model covering fiction, dialogue, roleplay, texts, emails, and notes
• EQ-Bench 4 score of 1330
• A repost notes its chat template supports image and video input (visual multimodality)

Unconfirmed

• Poster @Lukinator6446 claims the score beats GPT-5.5 and slightly trails Claude; this is a self-reported comparison without independent verification
• Reposters @CurieuxExplorer and @burnytech have not yet tested the model; whether it truly fixes homogeneous AI writing awaits community benchmarks

Why it matters

• The team's motivation is clear: the author felt ChatGPT stories read as formulaic, like a polite intern's work, and built a model with more human voice
• A tiny team entering the creative-writing niche with open weights offers a locally deployable option for fiction, roleplay, and everyday text tasks

2026-09-21 ~ 2026-09-22 · 7 related posts

• [source] Open-source Hemmingway-1: a 27B creative-writing model that beat GPT-5.5 in blind tests — Lukinator6446 · 2026-09-21
• [source] Two-Person Lab Open-Sources 27B Writing Model Hemmingway-1, 1330 on EQ-Bench 4 — lukinator644 · 2026-09-21
• Hemmingway-1, a Qwen3.8-27B fine-tune promising human-like writing, lands on Hugging Face — victormustar · 2026-09-21
• [source] Hemmingway-1: Qwen3.8-27B finetune that writes like a person — paf1138 · 2026-09-21
• Hemmingway-1, a Qwen3.8-27B-based model claiming human-like writing, draws skepticism — burny_tech · 2026-09-22

2 near-duplicate retellings: Lukinator6446 · CurieuxExplorer