오픈소스 자가 개선형 에이전트 코딩 모델 'Ornith-1.0' 공개

Ornith-1.0 Aloha! 🌺 Ornith-1.0 is a self-improving 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. Benchmarks Each model is evaluated against its size-appropriate baselines. All three use the same harnesses and decoding setup (see the notes under the tables). Ornith-1.0-9B Ornith-1.0-9B Qwen3.5-9B Qwen3.5-35B Gemma4-12B Gemma4-31B Agentic Coding Terminal-Bench 2.1 (Terminus-2) 43.1 21.3 41.4 21 42.1 Terminal-Bench 2.1 (Claude Code) 40.6 18.9 38.9 - - SWE-bench Verified 69.4 53.2 70 44.2 52 SWE-bench Pro 42.9 31.3 44.6 27.6 35.7 SWE-bench Multilingual 52 39.7 60.3 32.5 51.7 NL2Repo 27.2 16.2 20.5 10.3 15.5 Claw-eval Avg 63.1 53.2 65.4 32.5 48.5 SWE Atlas - QnA 17.9 9.2 13.2 - - SWE Atlas - RF 16.6 4.3 10.2 - - SWE Atlas - TW 15.3 4.4 9.8 - - Ornith-1.0-35B Ornith-1.0-35B Qwen3.5-35B Qwen3.6-35B Gemma4-31B Qwen3.5-397B Agentic Coding Terminal-Bench 2.1 (Terminus-2) 64.2 41.4 52.5 42.1 53.5 Terminal-Bench 2.1 (Claude Code) 62.8 38.9 49.2 - 48.6 SWE-bench Verified 75.6 70 73.4 52 76.4 SWE-bench Pro 50.4 44.6 49.5 35.7 51.6 SWE-bench Multilingual 69.3 60.3 67.2 51.7 69.3 NL2Repo 34.6 20.5 29.4 15.5 36.8 Claw-eval Avg 69.8 65.4 68.7 48.5 70.7 SWE Atlas - QnA 37.1 13.2 15.5 - 20.4 SWE Atlas - RF 29.7 10.2 11.4 - 18.4 SWE Atlas - TW 27.8 9.8 13.3 - 18.5 Ornith-1.0-397B 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): evaluated with the Harbor/Terminus-2 framework, parser=json, temperature=1.0, top_p=1.0, 128K context window. Each run uses a 4-hour timeout with 32 CPU cores and 48GB RAM, averaged over 5 runs. We adjust the Qwen chat template to keep training and inference consistent and modify Harbor to align with vLLM's reasoning_content key. * Terminal-Bench 2.1 (Claude Code): evaluated with Claude Code 2.1.126, parser=json, temperature=1.0, top_p=1.0, max_new_tokens=131072, averaged over 5 runs (Qwen chat template likewise modified). * SWE-bench Verified / Pro / Multilingual: OpenHands harness, temp=1.0, top_p=0.95, 256K context window. * SWE Atlas QnA / RF / TW: mini-SWE-agent harness, temp=1.0, top_p=0.95, 128K context window, averaged over 5 runs. * NL2Repo: temperature=1.0, top_p=1.0, 400K context, 48K output, anti-hacking filters. * ClawEval: an agentic code benchmark over real-user task distributions; temp=0.6, 256K context. Quickstart NOTE Ornith-1.0 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 requires recent runtimes: Transformers ≥ 5.8.1 vLLM ≥ 0.19.1 SGLang ≥ 0.5.9 Recommended sampling parameters: temperature=0.6 , top_p=0.95 , top_k=20 (use temperature=1.0 to reproduce the reported benchmark setup). Serving Ornith-1.0 Ornith-1.0 ships as a dense 9B model plus two Mixture-of-Experts models ( 35B , 397B ). All checkpoints expose the same OpenAI-compatible interface and support a 256K (262,144-token) context window ; the dense 9B fits on a single 80GB GPU, while the MoE checkpoints are sharded across a multi-GPU node with tensor parallelism. Each size is published in multiple precision / format variants: Checkpoint Architecture Format Best for Ornith-1.0-9B Dense (~9B) bf16 Single-GPU serving & fine-tuning Ornith-1.0-9B-GGUF Dense (~9B) GGUF (quantized) Local inference via llama.cpp / Ollama Ornith-1.0-35B MoE (35B) bf16 Full-precision multi-GPU serving Ornith-1.0-35B-FP8 MoE (35B) FP8 ~Half the VRAM on FP8-capable GPUs Ornith-1.0-35B-GGUF MoE (35B) GGUF (quantized) Local inference via llama.cpp / Ollama Ornith-1.0-397B MoE (397B) bf16 Full-precision serving on a multi-GPU node Ornith-1.0-397B-FP8 MoE (397B) FP8 Memory-efficient serving on FP8-capable GPUs The recipes below stand up an OpenAI-compatible server under the shared alias Ornith-1.0 . Set MODEL to the checkpoint you want, and match --tensor-parallel-size / --tp to your GPU count. vLLM # Pick a checkpoint — dense 9B, or MoE 35B / 397B (append -FP8 for lower-VRAM serving): MODEL=deepreinforce-ai/Ornith-1.0-397B # MoE checkpoints (35B / 397B): shard across the node with tensor parallelism. # Dense checkpoint (9B): fits on a single 80GB GPU — drop --tensor-parallel-size. vllm serve $MODEL \ --served-model-name Ornith-1.0 \ --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 # Pick a checkpoint — dense 9B, or MoE 35B / 397B (append -FP8 for lower-VRAM serving): MODEL=deepreinforce-ai/Ornith-1.0-397B # MoE checkpoints (35B / 397B): shard with --tp ; dense 9B: drop --tp for a single GPU. python -m sglang.launch_server \ --model-path $MODEL \ --served-model-name Ornith-1.0 \ --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 requires transformers >= 5.8.1 . The dense 9B checkpoint is the easiest to run locally. from transformers import AutoModelForCausalLM , AutoTokenizer model_name = "deepreinforce-ai/Ornith-1.0-9B" # or -35B / -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 : reaso