olka-fi/Ornith-1.0-397B-MXFP4

Ornith-1.0-397B — MXFP4 (mixed precision)

A 4-bit MXFP4 quantization of Ornith-1.0-397B, produced with qstream. Only the routed MoE experts (~95% of the weights) are quantized to MXFP4; everything quality-sensitive — including the always-on shared expert — stays BF16, bit-identical to the source.

The original model card follows in full below.

Size	225.9 GB (210 GiB weights), down from 793.6 GB BF16 source (28.5%, −72%)
Format	compressed-tensors mxfp4-pack-quantized — per-expert, FP4 E2M1, group-32 e8m0 scales, symmetric
Base	397B MoE · 512 experts (top-10) + 1 shared · hybrid attention (45 linear-attention / 15 full-attention) · 60 layers · hidden 4096 · vision encoder · post-trained on Qwen 3.5 for agentic coding

What is quantized

Component	Precision	Why
Routed experts (*mlp.experts.*proj)	MXFP4 (4-bit)	~95% of the weights, sparsely activated (top-10/512) — the only place worth the size win
Shared expert (*shared_expert.*)	BF16	dense, runs on every token — quantizing it roughly doubles PPL loss for <1% size, so kept lossless
Full + linear (SSM) attention, router gates	BF16	sensitive / per-token — kept lossless from the source
Vision encoder + projector	BF16	image path bit-identical to base
Embeddings, lm_head, norms	BF16	unchanged

The source ships the routed experts as fused 3D tensors (experts.gate_up_proj [512, 2048, 4096]); qstream splits them into per-expert 2D MXFP4 tensors that stock vLLM consumes. The config.json ignore list (995 entries) names every BF16 module.

Quality & faithfulness

Metric	Result	What it shows
WikiText-2 perplexity (test, 296 547 tokens)	5.21	language modeling intact — a broken quant lands in the hundreds
GSM8K (40-task sample, CoT)	39/40 (97.5%)	math/reasoning preserved (the 1 miss was length-truncated, not wrong reasoning)
Routed-expert SQNR (sampled, MXFP4 vs BF16)	≈ 19.0 dB (18.0–19.6 dB; 60 tensors across 5 layers × 4 experts × 3 proj)	reconstruction error is just the unavoidable 4-bit rounding

• Only the routed-expert FFN weights changed. ~95% of the weights are re-quantized to MXFP4; everything else is bit-identical BF16 to the source.
• Functional evals run on a single 268 GB GPU (NVIDIA B300): PPL 5.21 (well below the 35B sibling's 7.77, as expected for a far larger model) and near-perfect GSM8K confirm the quant is coherent end-to-end.
• SQNR ≈ 19 dB matches the 35B sibling (18.7 dB, where it corresponded to +0.94% WikiText-2 perplexity), and confirms the fused→per-expert split is correct (a wrong split would collapse SQNR toward 0 dB).

Serving with vLLM

The architecture (Qwen3_5MoeForConditionalGeneration) is registered natively in vLLM ≥ 0.20.0. At 225 GB it fits a single large-VRAM GPU (e.g. NVIDIA B300, 268 GB — verified) or multi-GPU tensor parallelism (e.g. 4× 80 GB). Single-GPU:

vllm serve olka-fi/Ornith-1.0-397B-MXFP4 \
  --max-model-len 8192 --gpu-memory-utilization 0.92 --max-num-seqs 128 \
  --load-format fastsafetensors --trust-remote-code
# multi-GPU: add --tensor-parallel-size N (and drop --max-num-seqs if VRAM is ample)
# image: vllm/vllm-openai:latest-cu130 (vLLM 0.20.0, CUDA 13.0) verified for the arch

On Blackwell the MXFP4 MoE runs through vLLM's MarlinExperts kernel as weight-only FP4 (W4A16): 4-bit weights dequantized, matmul in BF16 — full memory savings, lossless activations.

Note: config.json declares mtp_num_hidden_layers: 1, but the base release ships no MTP head weights, so multi-token speculative decoding is not available with this checkpoint.

How it was made

qstream-quantize \
  --model_dir <deepreinforce-ai/Ornith-1.0-397B> \
  --output_dir ./Ornith-1.0-397B-MXFP4 \
  --include_layers '*mlp.experts.*proj' \
  --device cuda --workers 6

--include_layers is an allowlist (routed experts only); the shared expert, attention, SSM, vision, routers, and embeddings all pass through as BF16.

License

Inherits the MIT license from the base model. This is a derivative (quantized) work of Ornith-1.0-397B.

---

Original model card

Ornith-1.0-397B

Aloha! 🌺 Today, we are releasing Ornith-1.0, a self-improving family of open-source models for agentic coding.

Highlights:

• State-of-the-Art Coding Agents: Available in 9B-Dense, 31B-Dense, 35B-MoE, and 397B-MoE (post-trained on top of Gemma 4 and Qwen 3.5), achieving state-of-the-art performance among open-source models of comparable size on coding benchmarks such as Terminal-Bench 2.1, SWE-Bench, NL2Repo and OpenClaw.
• Self-Improving Training Framework: Ornith-1.0 employs RL to learn to generate not only solution rollouts, but also the scallfold that drive those rollouts. By jointly optimizing the scaffold and the resulting solution, the model discovers better search trajectories and generates higher-quality solutions.
• Licence: MIT licensed, globally accessible, and free from regional limitations.

Ornith 1.0 397B

This model card documents Ornith-1.0-397B, the lightweight member of the Ornith family, designed for efficient single-GPU deployment.

Benchmarks

	Ornith-1.0-397B	Qwen3.5-397B	Qwen3.7-Max	GLM-5.2-744B	Minimax-M3-428B	DeepSeek-V4-Pro-1.6T	Claude Opus 4.7	Claude Opus 4.8
Agentic Coding
Terminal-Bench 2.1 (Terminus-2)	77.5	53.5	73.5	81.0	64	64	70.3	85
Terminal-Bench 2.1 (Claude Code)	78.2	48.6	69.8	82.7	-	66.5	69.7	78.9
SWE-bench Verified	82.4	76.4	80.4	-	-	80.6	80.8	87.6
SWE-bench Pro	62.2	51.6	60.6	62.1	59	55.4	64.3	69.2
SWE-bench Multilingual	78.9	69.3	78.3	-	-	76.2	-	-
NL2Repo	48.2	36.8	47.2	48.9	42.1	-	-	69.7
Claw-eval Avg	77.1	70.7	65.2	-	-	75.8	78.2	-
SWE Atlas - QnA	41.2	20.4	-	-	37.9	27.2	40.3	48.8
SWE Atlas - RF	42.6	18.4	-	-	-	-	48.6	46.7
SWE Atlas - TW	39.1	18.5	-	-	30.8	-	38.5	-

* Terminal-Bench 2.1 (Terminus-2): We evaluate Terminal-Bench 2.1 using the Harbor/Terminus-2 framework with parser=json, temperature=1.0, top_p=1.0, and a 128K context window. Each run uses a 4-hour timeout with 32 CPU cores and 48GB RAM, and results are averaged over 5 runs. We adjust the Qwen chat template to ensure consistency between training and inference (https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B/blob/main/chat_template.jinja), and modify Harbor to align with vLLM's reasoning_content key.
* Terminal-Bench 2.1 (Claude Code): We evaluate Terminal-Bench 2.1 using Claude Code 2.1.126 with parser=json, temperature=1.0, top_p=1.0, max_new_tokens=131072. Results are averaged over 5 runs. Again, Qwen chat template needs to be modified.
* SWE-Bench Verified, Pro and Multilingual: using OpenHands harness with temp=1.0, top_p=0.95, 256k context window.
* SWE Atlas QnA, RF, TW: using mini SWE agent harness with temp=1.0, top_p=0.95, 128K context window. Results are averaged over 5 runs.
* NL2Repo: with temperature=1.0, top_p=1.0, 400K context, 48K output and anti-hacking filters.
* ClawEval: An agentic code benchmark over real-user task distributions; temp=0.6 and 256K context.

Quickstart

📝 NOTE

Ornith-1.0-397B is a reasoning model: by default the assistant turn opens with a <think> … </think> block before the final answer. The serving recipes below enable a reasoning parser so the chain-of-thought is returned in a separate reasoning_content field, and a tool-call parser so the model's <tool_call> blocks are surfaced as OpenAI-style tool_calls.

Serving Ornith-1.0-397B requires recent runtimes:

• Transformers ≥ 5.8.1
• vLLM ≥ 0.19.1
• SGLang ≥ 0.5.9

Serving Ornith-1.0-397B

The two recipes below stand up an OpenAI-compatible server on a single 8×80GB GPU node (tensor-parallel 8). Adjust --tensor-parallel-size / --tp to the number of GPUs you have.

vLLM

vllm serve deepreinforce-ai/Ornith-1.0-397B \
    --served-model-name Ornith-1.0-397B \
    --tensor-parallel-size 8 \
    --host 0.0.0.0 --port 8000 \
    --max-model-len 262144 \
    --gpu-memory-utilization 0.90 \
    --enable-prefix-caching \
    --enable-auto-tool-choice --tool-call-parser qwen3_xml \
    --reasoning-parser qwen3 \
    --trust-remote-code

SGLang

python -m sglang.launch_server \
    --model-path deepreinforce-ai/Ornith-1.0-397B \
    --served-model-name Ornith-1.0-397B \
    --tp 8 \
    --host 0.0.0.0 --port 8000 \
    --context-length 262144 \
    --mem-fraction-static 0.85 \
    --tool-call-parser qwen3_coder \
    --reasoning-parser qwen3

Hugging Face Transformers

For a quick local test (or to script offline generation), load the model directly with Transformers. Make sure you have a recent release installed — see the Transformers installation guide; Ornith-1.0-397B requires transformers >= 5.8.1.

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "deepreinforce-ai/Ornith-1.0-397B"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    dtype="auto",
    device_map="auto",
)

messages = [
    {"role": "user", "content": "Write a Python function is_prime(n). Keep it short."}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)

inputs = tokenizer(text, return_tensors="pt").to(model.device)
generated = model.generate(
    **inputs,
    max_new_tokens=512,
    do_sample=True,
    temperature=0.6,
    top_p=0.95,
    top_k=20,
)
output_ids = generated[0][inputs.input_ids.shape[1]:]

# The reply contains a <think> ... </think> reasoning block followed by the answer.
content = tokenizer.decode(output_ids, skip_special_tokens=True)
print(content)

To split the reasoning trace from the final answer, parse on the </think> marker:

text = tokenizer.decode(output_ids, skip_special_tokens=True)
if "</think>" in text:
    reasoning, answer = text.split("</think>", 1)
    reasoning = reasoning.replace("<think>", "").strip()
    answer = answer.strip()
else:
    reasoning, answer = "", text.strip()

Using Ornith-1.0-397B via the Chat Completions API

Once a vLLM or SGLang server is running, talk to it with any OpenAI-compatible client.

Basic Usage

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:8000/v1",
    api_key="EMPTY",  # any non-empty string works for a local server
)

response = client.chat.completions.create(
    model="Ornith-1.0-397B",
    messages=[
        {"role": "user", "content": "Write a one-line Python lambda that squares a number."}
    ],
    temperature=0.6,
    top_p=0.95,
    max_tokens=1024,
)

message = response.choices[0].message
# reasoning_content holds the <think> trace; content holds the final answer.
print("reasoning:", getattr(message, "reasoning_content", None))
print("answer:", message.content)

You can also stream tokens, or hand the model tools — Ornith-1.0-397B emits well-formed function calls that the server parses into the standard tool_calls field:

tools = [
    {
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "Get the current weather for a city",
            "parameters": {
                "type": "object",
                "properties": {"city": {"type": "string"}},
                "required": ["city"],
            },
        },
    }
]

response = client.chat.completions.create(
    model="Ornith-1.0-397B",
    messages=[{"role": "user", "content": "What is the weather in Paris right now?"}],
    tools=tools,
    tool_choice="auto",
    temperature=0.6,
    max_tokens=2048,
)

tool_call = response.choices[0].message.tool_calls[0]
print(tool_call.function.name, tool_call.function.arguments)
# -> get_weather {"city": "Paris"}

You can point any OpenAI-compatible SDK (Python, Node.js, etc.) or curl at the same /v1/chat/completions endpoint.

Agentic Usage

Ornith-1.0-397B excels in tool-calling and agentic coding capabilities.

Agent Frameworks

Because Ornith-1.0-397B exposes an OpenAI-compatible endpoint with tool calling, it works out of the box with standard agent frameworks. Below is a minimal example that connects Ornith-1.0-397B to tools through an MCP server.

import os
from openai import OpenAI

client = OpenAI(
    base_url=os.getenv("OPENAI_BASE_URL", "http://localhost:8000/v1"),
    api_key=os.getenv("OPENAI_API_KEY", "EMPTY"),
)

tools = [
    {
        "type": "function",
        "function": {
            "name": "run_shell",
            "description": "Run a shell command and return its output.",
            "parameters": {
                "type": "object",
                "properties": {
                    "command": {"type": "string", "description": "The command to run"}
                },
                "required": ["command"],
            },
        },
    }
]

messages = [{"role": "user", "content": "List the Python files in the current directory."}]

response = client.chat.completions.create(
    model="deepreinforce-ai/Ornith-1.0-397B",
    messages=messages,
    tools=tools,
    temperature=0.6,
    top_p=0.95,
)
print(response.choices[0].message)

Examples of using Ornith with agent harness:

Hermes Agent

# Hermes talks to any OpenAI-compatible endpoint — point it at your Ornith server.
export OPENAI_BASE_URL="http://localhost:8000/v1"
export OPENAI_API_KEY="EMPTY"
export MODEL="deepreinforce-ai/Ornith-1.0-397B"

OpenClaw

# OpenClaw talks to any OpenAI-compatible endpoint — point it at your Ornith server.
export OPENAI_BASE_URL="http://localhost:8000/v1"
export OPENAI_API_KEY="EMPTY"
export OPENAI_MODEL="deepreinforce-ai/Ornith-1.0-397B"

Unsloth Studio

pip install unsloth

# Load Ornith for fast local inference or fine-tuning (Python):
#   from unsloth import FastLanguageModel
#   model, tokenizer = FastLanguageModel.from_pretrained(
#       "deepreinforce-ai/Ornith-1.0-397B",
#       max_seq_length=262144,
#       load_in_4bit=True,
#   )

OpenHands

pip install openhands-ai

# OpenHands routes through LiteLLM; the "openai/" prefix selects the OpenAI-compatible path.
export LLM_MODEL="openai/deepreinforce-ai/Ornith-1.0-397B"
export LLM_BASE_URL="http://localhost:8000/v1"
export LLM_API_KEY="EMPTY"

# Launch the CLI (or run the official OpenHands Docker image with the same env vars).
openhands

Coding CLIs

Ornith-1.0-397B is optimized for terminal-based coding agents. Point any OpenAI-compatible coding CLI at your Ornith-1.0-397B endpoint (set OPENAI_BASE_URL and OPENAI_API_KEY) to understand large codebases, automate tedious work, and ship faster.

OpenCode

# Register your local Ornith endpoint as a provider in ~/.config/opencode/opencode.json:
#
# {
#   "$schema": "https://opencode.ai/config.json",
#   "provider": {
#     "ornith": {
#       "npm": "@ai-sdk/openai-compatible",
#       "name": "Ornith (local)",
#       "options": { "baseURL": "http://localhost:8000/v1", "apiKey": "EMPTY" },
#       "models": { "deepreinforce-ai/Ornith-1.0-397B": { "name": "Ornith-1.0-397B" } }
#     }
#   }
# }

opencode

Citation

If you find our work helpful, feel free to give us a cite.

@misc{ornith_397b,
    title = {{Ornith-1.0-397B}: Agentic Coding, Open to All},
    url = {https://deep-reinforce.com/ornith_1_0.html},
    author = {{DeepReinforce Team}},
    year = {2026}
}