DeepReinforce

Ornith 1.0 9B

Name: Ornith 1.0 9B
Rating: 77 (185 reviews)
Author: DeepReinforce

前沿

HuggingFace

1.6M下载量487点赞Jun 2026发布日期262K tokens上下文MIT许可证66 良好质量

Ornith 1.0 9B (9.399999618530273B parameters) requires approximately 7.7 GB of VRAM with Q4_K_M quantization. For the best balance of quality and speed, we recommend hardware with at least 9 GB of VRAM.

快速开始

— 复制粘贴即可本地运行

Copy-paste commands to run Ornith 1.0 9B on your machine.

Run

docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
  --hf-repo "deepreinforce-ai/Ornith-1.0-9B" \
  --hf-file "Ornith-1.0-9B-Q4_K_M.gguf" \
  -c 4096 -ngl 99

Quick specs

Parameters9.4B

Architecturedense

Context262K tokens

Modalitytext

Min RAM3.7 GB

Rec. RAM5.7 GB (Q4_K_M)

LicenseMIT

FamilyOrnith

✓ Code✓ Reasoning

About this model

Ornith-1.0-9B is the most lightweight member of DeepReinforce's self-improving Ornith family of open coding agents, post-trained on top of Qwen 3.5. At 9.4B dense parameters with a 262K context and a hybrid linear/full-attention backbone, it is designed for efficient single-GPU agentic coding, and is MIT licensed with no regional restrictions.

•9.4B dense — the lightest Ornith, sized for a single consumer GPU.
•Self-improving RL: jointly optimizes solution rollouts and the scaffolds that drive them.
•SOTA among comparable open models on Terminal-Bench 2.1, SWE-Bench, NL2Repo and OpenClaw.
•262K context with hybrid linear + full attention (1-in-4 layers).
•MIT licensed, globally accessible.

Ornith 1.0 9B 的最佳选择

运行此模型

Ornith 1.0 9B on RTX 3080 Ti 12GB Ornith 1.0 9B on RTX 3080 12GB Ornith 1.0 9B on RTX 5070 12GB

量化选项

各量化级别的 VRAM 估算

No hardware detected — fit column shows raw VRAM estimates

Quant	Bits	VRAM	Quality	Fit
Q1_0_G128	1.125	1.4 GB	Very Low	—
Q2_0_G128	1.71	2.5 GB	Low	—
Q2_K	2	3.7 GB	Low	—
Q3_K_S	3	4.6 GB	Low	—
NVFP4	4	5.3 GB	Medium	—
Q4_K_M	4	5.7 GB	Medium	—
Q5_K_M	5	6.8 GB	High	—
Q6_K	6	7.7 GB	High	—
Q8_0	8	10.1 GB	Very High	—
F16	16	19.3 GB	Maximum	—

Quality benchmarks

Ornith 1.0 9B benchmark scores

Benchmark verified

Coding

SWE-bench Verified69.4%

HumanEval+—

Aider Polyglot—

LiveCodeBench—

Source: official · 2026-06-21

硬件兼容性

全部硬件的适配估算

打开计算器

Computing compatibility...

内存详细分析

Reference: RTX 2060 6GB

Weights5.7 GB

KV Cache0.5 GB

Runtime0.9 GB

Headroom0.6 GB

常见问题

FAQ — Ornith 1.0 9B

另请参阅

量化指南评分方法打开计算器