Altworld · 2026-09-20 · major
Hemmingway-1 — a 27B open model tuned for everyday messages
Hemmingway-1 is a 27B open-weight model from Altworld, fine-tuned from Qwen3.8-27B for the messages and emails people write every day. It ships under Apache-2.0 and placed third on the independent EQ-Bench 4 leaderboard.
A 27B open model that answers a request for an email with the email, and nothing else.
Quick facts
| Maker | Altworld |
|---|---|
| Parameters | 27B |
| Built on | Qwen3.8-27B |
| License | Apache-2.0, commercial use allowed |
| Context window | 262,144 tokens |
| API model id | hemmingway-27b |
| Availability | Open weights on Hugging Face + hosted API |
Benchmarks
| Fable 5 | 1341 | |
|---|---|---|
| Kimi K3 | 1332 | |
| Hemmingway-1 | 1330 | |
| GPT-5.5 | 1316 | |
| Claude Opus 4.7 | 1312 | |
| Claude Opus 4.8 | 1285 |
| Hemmingway-1 | 1026 | |
|---|---|---|
| Fable 5.1 | 1024 | |
| Fable 5 | 1009 | |
| GLM-5.3 | 1007 | |
| Kimi K3 | 996 | |
| GPT-6 Astra | 976 | |
| Qwen3.8-27B base | 954 |
Pricing
| Input | $0.24 / 1M tokens |
|---|---|
| Cached input | $0.024 / 1M tokens |
| Output · Thinking tokens included | $0.90 / 1M tokens |
| Free tier · $5 minimum prepaid credit | None |
What is it?
Hemmingway-1 hands back the message you asked for instead of three options, a preamble and a paragraph explaining the options. Altworld trained the 27B open-weight model on the writing people do all the time — the note to a landlord, the reply to a coworker, the message that keeps coming out wrong. Weights are on Hugging Face under Apache-2.0, and the same model is served through Altworld's own apps and API.
How does it work?
Underneath is a fine-tune of Qwen3.8-27B, with a 262,144-token context window and the standard Transformers and vLLM paths for running it yourself. Altworld scored it by blind head-to-head matchups on eighty real requests: each answer was paired against another model's answer to the same request, run in both orders so position could not sway the result, and judged by a model that was not among those being compared.
Why does it matter?
Three of the four scoreboards behind those wins are Altworld's own, and the company says so on the model card. The one that is not — EQ-Bench 4, run by its own public harness — put Hemmingway-1 third, ahead of GPT-5.5, Claude Opus 4.7 and Claude Opus 4.8, and twelve points off the top. A 27B model landing there means a machine with a single consumer GPU can cover routine message drafting without an API bill.
Who is it for?
people who draft a lot of routine messages, and anyone running writing models locally
Frequently asked questions
- How much does the Hemmingway-1 API cost?
- Altworld charges $0.24 per million input tokens, $0.024 per million cached input tokens, and $0.90 per million output tokens, with thinking tokens counted as output. There is no free tier — the platform requires $5 of prepaid credit before the first call. Accounts on credit get eight concurrent requests and a 32,768-token ceiling per request.
- Are Hemmingway-1's benchmark wins independently verified?
- Only one of the four is. Altworld states on the model card that CommunicationBench, Human-Likeness and StoryBench are benchmarks it built and ran itself. EQ-Bench 4 is a public emotional-intelligence benchmark run by its own harness, and Hemmingway-1 placed third on it. Altworld does say every in-house matchup was blind, run in both orders, and judged by a model outside the comparison.
- Where does Hemmingway-1 do worse than bigger models?
- Long and hostile fiction. Altworld reports that Hemmingway-1 scores 1197 on its StoryBench, level with Kimi K3 but behind Fable 5 Max at 1277 and GLM-5.3 at 1254, and says the dedicated story models are simply better at extended narrative turns. Its wins concentrate in money and admin, workplace messages, difficult asks and persuasion.
- Can I use Hemmingway-1 commercially?
- Yes. Hemmingway-1 ships under Apache-2.0, which Altworld describes on the model card as yours to use including commercially. The full BF16 weights are downloadable from Hugging Face, so you can host it privately, and community quantizations in GGUF and MLX formats already exist for smaller machines.
- What should Hemmingway-1 not be used for?
- Altworld's own fine print rules out medical, legal and financial decisions, warning that Hemmingway-1 can be wrong and still sound certain. The model is also English-first, so other languages are outside what it was tuned and measured on.
Try it
vllm serve Altworld/Hemmingway-1 --max-model-len 262144