whatmodel.sh / Arete-27B

Arete-27B is an open-weights language model from Nous Research, released November 2026. Its published size is 27B parameters. Licence: apache-2.0.

Catalogue identity notes

Nous Research SFT-only fine-tune of Qwen/Qwen3.8-27B (card base_model), 'post-trained ... to reinforce western values', teacher Inkling, text-only data, vision encoder frozen, MTP head disabled in training; evaluated checkpoint step 1,202. Repo created 2026-11-06. Training context 64k; the card states no inference context length, so none recorded. Intended 'for general assistance and agentic work where users want this change in viewpoint'. No official Ollama entry (404).

Model information sources

Release date ↗ · Weights licence ↗ · Parameter count ↗

Dense. Card 'Parameter scale | 27B'. HF safetensors metadata at 108f0af: 27,356,728,560 BF16 parameters (includes the frozen vision encoder).

No comparable score yet. Its published results are below, but they don't overlap enough with other models to calculate a score.

$ how it compares

Scored LLMs released closest to Arete-27B, out of 1206 with a score, then more from Nous Research.

#modelscorewhereRAMreleased
1Claude Fable 5.2 Anthropic89±1api $16.00/M2026-11
14Gemini 4 Hydrogen Google79±2api $3.00/M2026-11
22Hy4 Tencent78±1api price unknown2026-11
32Muse Ember Meta76±1api $0.20/M2026-11
37Inkling 2 Small (594B-A17B) Thinking Machines75±1api price unknown2026-11
more from Nous Research
511Hermes 3 Llama 3.1 405B Nous Research53±1local~114GB2024-08
614Nous Hermes 2 Yi 34B Nous Research50±6local~22GB2023-12
616Hermes 3 Llama 3.1 70B Nous Research50±1local~43GB2024-07

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasksnot ranked on this task yet · 1 test
testresultvs the others in its rundifficultycountswhy
Terminal-Bench 2.1 lab70.0% ✎▁·▁·▁▁▄▃▅█▆█ beats 38/68—not counted yet: no other model in the same run to compare it with

$ where to run it

Memory is what a version takes in GPU memory (or a Mac's unified memory) with an 8k-token conversation. With less, it runs partly from system RAM, more slowly. Enter your machine to see what runs at full speed.

llama.cpp · recommended

version

  1. Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.
    macOS / Linux
    curl -LsSf https://llama.app/install.sh | sh
    Windows
    irm https://llama.app/install.ps1 | iex

    other ways to install ↗

  2. Run Arete-27B · needs 20.5 GB memory, 18.1 GB download · from huggingface.co/bartowski
    chat in the terminal
    llama cli -hf bartowski/NousResearch_Arete-27B-GGUF:Q4_K_M
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf bartowski/NousResearch_Arete-27B-GGUF:Q4_K_M

LM Studio

version

  1. Install LM Studio · Open the app once before using the lms command.

    download it (macOS / Windows / Linux) ↗

  2. Run Arete-27B · needs 20.5 GB memory, 18.1 GB download · from huggingface.co/bartowski

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/bartowski/NousResearch_Arete-27B-GGUF@Q4_K_M -y
  4. Then run lms ls to see its name, and lms chat followed by that name

Ollama

version

  1. Install Ollama

    download it (macOS / Windows) ↗

    Linux
    curl -fsSL https://ollama.com/install.sh | sh
  2. Run Arete-27B · needs 20.5 GB memory, 18.1 GB download · from huggingface.co/bartowski
    chat in the terminal
    ollama run hf.co/bartowski/NousResearch_Arete-27B-GGUF:Q4_K_M

Model repository and documentation: huggingface.co/NousResearch ↗

$ published results · 1 result, not enough overlap for a score
testresultsource and conditions
Terminal-Bench 2.170.0%NousResearch/Arete-27B model card · maker-reported

Nous Research own run in the pinned NousResearch/Arete-27B card (sha 108f0af), 'General and agentic capabilities' table, row 'Terminal-Bench-2.1', column 'Arete-27B' = 70%. Card: 'We ran our model and Qwen3.8-27B with a like-for-like configuration on our hardware ... We do not take credit for any seeming improvement in the Terminal-Bench-2.1 score.' Harness and thinking mode not stated. Qwen3.8-27B column not used.

$ improve this page

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json