Instructions to use wsgpolar/Anura-R1-Ultra with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use wsgpolar/Anura-R1-Ultra with MLX:

# Make sure mlx-vlm is installed
# pip install --upgrade mlx-vlm

from mlx_vlm import load, generate
from mlx_vlm.prompt_utils import apply_chat_template
from mlx_vlm.utils import load_config

# Load the model
model, processor = load("wsgpolar/Anura-R1-Ultra")
config = load_config("wsgpolar/Anura-R1-Ultra")

# Prepare input
image = ["http://images.cocodataset.org/val2017/000000039769.jpg"]
prompt = "Describe this image."

# Apply chat template
formatted_prompt = apply_chat_template(
    processor, config, prompt, num_images=1
)

# Generate output
output = generate(model, processor, formatted_prompt, image)
print(output)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use wsgpolar/Anura-R1-Ultra with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "wsgpolar/Anura-R1-Ultra"

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "wsgpolar/Anura-R1-Ultra"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent

How to use wsgpolar/Anura-R1-Ultra with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "wsgpolar/Anura-R1-Ultra"

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default wsgpolar/Anura-R1-Ultra

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use wsgpolar/Anura-R1-Ultra with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "wsgpolar/Anura-R1-Ultra"

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "wsgpolar/Anura-R1-Ultra" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Anura R1 Ultra

Anura R1 Ultra is an agentic coding model by Genysis IQ (version 1.0.0).

Agentic software engineering: reads and edits code, runs commands and tests, plans multi-step work
Web search and crawling for current information, with cited sources
Vision: screenshots, diagrams, UI mockups, charts and photos
Fast local inference on Apple Silicon with MLX (mixture-of-experts, ~3B active parameters per token)
128K-token context in the default engine configuration

Results (measured on an Apple M5, 32 GB)

	Anura R1 Ultra
Identity, held-out questions (Anura system prompt / served with identity preamble)	26/26 · 26/26
Identity, bare weights under a generic third-party prompt	24/26
Vision (shapes, OCR, code screenshot, chart, UI→HTML)	5/5
Agent coding tasks, pass@1, 40 verified tasks in 6 languages	36/40
Decode speed	~48 tok/s
Cached follow-up prefill (14.7K-token prompt)	0.18 s

The coding tasks are the suite the training trajectories came from, so they show the fine-tune kept the foundation's skills; they are not a held-out benchmark.

Format

MLX 4-bit weights plus the Anura R1 Ultra LoRA adapter in anura_adapter/, applied at load time. The chat template is Anura R1 Ultra's own (chat_template.jinja). About 20 GB of memory is needed (32 GB Apple Silicon Mac recommended).

Running it

The intended runtime is the Anura engine and CLI from Genysis IQ (anura serve, anura), which adds the agent tools, sessions, web UI and an OpenAI-compatible API. To load the weights directly with mlx-vlm:

from mlx_vlm import load, generate

model, processor = load("<this repo>", adapter_path="<this repo>/anura_adapter")

License and provenance

Access to this repository is limited to licensed users of Genysis IQ. Genysis IQ built Anura R1 Ultra by fine-tuning an open-weight foundation model released under the Apache License 2.0; those foundation weights remain under that license, and NOTICE gives the attribution it requires. Genysis IQ's adapter and packaging are provided under Genysis IQ's commercial terms.

Downloads last month: 8

Safetensors

Model size

35B params

Tensor type

U32

BF16

MLX

Hardware compatibility

4-bit