Hemmingway-1 is a 27B-parameter open-weights model from Altworld, fine-tuned from Qwen/Qwen3.8-27B and aimed at the everyday writing people actually do: messages to a landlord, emails, the awkward note to a colleague. Its stated design goal is to return the text itself rather than a preamble, a list of options and commentary around it. The weights ship in bfloat16 under the Qwen3_5ForCausalLM architecture, a 64-layer hybrid stack with a 5,120 hidden size in which every fourth layer uses full attention and the rest use linear attention, with 24 attention heads, 4 key-value heads, a 248,320-token vocabulary and a 262,144-token context window. It loads through Transformers, and the card gives a vLLM command that serves it with --max-model-len 262144. The model is English-first, and the card warns that it can be wrong while sounding certain, so it should not be used for medical, legal or financial decisions. Altworld reports results from its own CommunicationBench, Human-Likeness and StoryBench evaluations, which it says were blind, run in both orders, and judged by a different model from those being compared; the CommunicationBench set is eighty real requests. The developer states that it ranks first on these in-house tests against larger proprietary and open models, so those comparisons should be read as vendor-run rather than independent. The one external result cited is EQ-Bench 4, the public emotional-intelligence benchmark run by its own harness, where Altworld says the model placed third. Licensing matters for adopters: the weights are released under CC BY-NC 4.0, free for personal, research and other non-commercial use including fine-tuning and sharing with credit, while commercial use requires a separate agreement with Altworld. A companion code repository is published on GitHub at lukeckprobierts/Hemmingway-1, and Altworld also offers a hosted web app and Mac and Android apps.