Releases·Model Release
Altworld open-sources Hemmingway-1, an Apache-2.0 27B creative-writing fine-tune of Qwen3.8-27B scoring EQ-Bench 4 1330
Altworld, a small lab based in Switzerland and South Africa, open-sourced Hemmingway-1, a 27B fine-tune of Qwen3.8-27B specialized for creative writing and short-form personal text. The Apache-2.0 weights are on Hugging Face and the model is vLLM-compatible. It posts EQ-Bench 4 1330 and tops the lab's internal blind pairwise writing benchmarks against the frontier models tested.
87 engagement2 sourcesSun, Sep 20, 2026, 07:54 PM
AI Overview
- Switzerland/South Africa lab Altworld released 54.7 GB bf16 weights on Hugging Face, vLLM-compatible with MTP layer included
- Tops Altworld's internal blind pairwise benchmarks: CommunicationBench 1026 and human-likeness 1032 against frontier models, LLM-judged in both presentation orders
- Targets creative writing and short-form personal text (stories, dialogue, roleplay, emails); math, code, and factual recall unchanged from base
- Altworld flags its internal rankings as its own LLM-judged benchmarks with standard caveats; demo at hemmingway.io
Related
Developments·Benchmark
Altworld's Hemmingway-1 (27B Qwen3.8 fine-tune) hits ~32 tok/s on M5 Max, excels at role-play fiction — llm-bench.io
A tester ran Altworld's Hemmingway-1, a 27B fine-tune of Qwen3.8-27B built for human-like writing, on a MacBook M5 Max at ~31.8 tok/s with thinking and MTP enabled. An LLM judge scored it 94.2 on role play and narrative, narrowly below one Qwen3.8-27B run (95.35), but side-by-side story samples read noticeably more like human fiction. The verdict: strong pick for fiction and everyday writing, while base Qwen remains the better generalist or coding assistant.
A tester ran Altworld's Hemmingway-1, a 27B fine-tune of Qwen3.8-27B built for human-like writing, on a MacBook M5 Max at ~31.8 tok/s with thinking and MTP enabled. An LLM judge scored it 94.2 on role play and narrative, narrowly below one Qwen3.8-27B run (95.35), but side-by-side story samples read noticeably more like human fiction. The verdict: strong pick for fiction and everyday writing, while base Qwen remains the better generalist or coding assistant.
5 engagement1 sourceredditThu, Oct 1, 2026, 07:28 PM
Social·Community
User shares custom MXFP6/MXFP8 quant for Qwen3.8-27B on RTX Pro 6000
A user released a custom quantized version of Qwen3.8-27B, optimized for RTX Pro 6000 cards, using a mixed MXFP6/MXFP8 format. The quant prioritizes quality while improving speed, and requires a custom llama fork. It supports 256k context and is available on Hugging Face.
A user released a custom quantized version of Qwen3.8-27B, optimized for RTX Pro 6000 cards, using a mixed MXFP6/MXFP8 format. The quant prioritizes quality while improving speed, and requires a custom llama fork. It supports 256k context and is available on Hugging Face.
8 engagement1 sourceredditSun, Aug 23, 2026, 02:27 AM
Releases·Model Release
Qwen3.8-27B now runs on Cerebras with rapid inference
Qwen3.8-27B is now available on Cerebras' inference platform, announced via an X post promoting its fast, open-weight serving. This gives developers access to the open Qwen model with Cerebras' high-speed inference.
Qwen3.8-27B is now available on Cerebras' inference platform, announced via an X post promoting its fast, open-weight serving. This gives developers access to the open Qwen model with Cerebras' high-speed inference.
1.0k engagement1 sourcexSat, Sep 12, 2026, 07:35 AM
Social·Community
Reddit AMA opens on writing benchmark tracking 167 models' creative-writing capabilities
A Reddit user opened an AMA about a creative writing benchmark that tracks 167 models on a pareto frontier of capability versus cost. The host offers to answer questions about specific models, costs, tasks, and performance. The thread gives practitioners a venue to probe how models compare on creative writing tasks.
A Reddit user opened an AMA about a creative writing benchmark that tracks 167 models on a pareto frontier of capability versus cost. The host offers to answer questions about specific models, costs, tasks, and performance. The thread gives practitioners a venue to probe how models compare on creative writing tasks.
7 engagement1 sourceredditSat, Sep 26, 2026, 07:59 PM
Social·Community
SWE builds 3x RTX 3090, 64GB VRAM local AI coding PC running Qwen3.8-27B via llama.cpp
A software engineer detailed a DIY workstation with three RTX 3090s (~62GB usable VRAM) built to replace paid OpenAI/Anthropic subscriptions for coding agents. The build runs a Qwen3.8-27B Q8_0 GGUF quant through a self-compiled llama.cpp server with 262k context, at 44-46 t/s dropping to 28-30 t/s as context fills. It shows local multi-GPU setups are now practical for real software engineering workflows, though with nontrivial hardware constraints around PSUs, risers, and cooling.
A software engineer detailed a DIY workstation with three RTX 3090s (~62GB usable VRAM) built to replace paid OpenAI/Anthropic subscriptions for coding agents. The build runs a Qwen3.8-27B Q8_0 GGUF quant through a self-compiled llama.cpp server with 262k context, at 44-46 t/s dropping to 28-30 t/s as context fills. It shows local multi-GPU setups are now practical for real software engineering workflows, though with nontrivial hardware constraints around PSUs, risers, and cooling.
70 engagement1 sourceredditSun, Sep 13, 2026, 02:16 PM