MODEL · DEEPREINFORCE AI · 9B DENSE / 31B DENSE / 35B-A3B MOE / 397B MOE — POST-TRAINED ON GEMMA 4 + QWEN 3.5 BASES
Ornith-1.0 (9B / 31B / 35B-MoE / 397B-MoE)
A self-improving open-source family of agentic coding models, post-trained on Gemma 4 and Qwen 3.5 bases. DeepReinforce reports SOTA among comparable open models on Terminal-Bench 2.1, SWE-Bench (Verified/Pro/Multilingual), NL2Repo, and OpenClaw. The training idea: RL that jointly optimizes both the solution rollout and the scaffold that drives it. MIT, no regional limits. The 9B / 31B / 35B are single-GPU-deployable; the 397B MoE is hosted / big-iron.
License: MIT · Context: 128K–400K depending on variant · Released: June 21, 2026
The decision in five lines
- The call
- Consider — runnable locally, family reference
- Best for
- Local evaluation and family reference
- Runs on
- 23 hardware picks fit (cheapest: Intel Arc B580 12 GB · $249)
- Watch out
- General chat or non-coding work — Ornith is tuned specifically for agentic coding loops.
- Evidence
- Estimated
- 9B dense
- PARAMETERS
- AGENTIC CODING MODELS
- TYPE
- 128K–400K
- CONTEXT
- ~5–6 GB (9B) / ~18 GB (31B / 35B-A3B) at Q4 — 397B is big-iron
- VRAM AT Q4
Where we recommend this
This model isn’t currently in an active planner slot. See the runner notes below if you’re running it anyway.
The call
A self-improving open-source family of agentic coding models, post-trained on Gemma 4 and Qwen 3.5 bases. DeepReinforce reports SOTA among comparable open models on Terminal-Bench 2.1, SWE-Bench (Verified/Pro/Multilingual), NL2Repo, and OpenClaw. The training idea: RL that jointly optimizes both the solution rollout and the scaffold that drives it. MIT, no regional limits. The 9B / 31B / 35B are single-GPU-deployable; the 397B MoE is hosted / big-iron.
When not to use: General chat or non-coding work — Ornith is tuned specifically for agentic coding loops. Benchmarks are vendor-reported; verify on your own repo before switching from Qwen3-Coder-30B-A3B or North Mini Code.
Runner notes
vLLM with a modified Qwen chat template (see the model card). 35B-A3B fits a single 24 GB rig at Q4; the 9B runs on ~8 GB. Community GGUFs emerging. Pair with a real agent harness (OpenHands, Claude Code, mini-SWE-agent) — it targets those loops.
Hardware that fits
Every hardware pick whose memory fits this model at the quant we recommend. Sorted cheapest-first — the top row is your best-value fit. Click through for the full buyer’s guide.
- Intel Arc B580 12 GBPerfect · 1.9× 12 GB · $249–$299
- NVIDIA RTX 3060 12 GBPerfect · 1.9× 12 GB · $280–$400
- Minisforum UM890 ProPerfect · 3.9× 32 GB DDR5 (shared) · $463–$580 all-in
- RTX 5060 Ti 16 GBPerfect · 2.6× 16 GB · $560–$610
- AMD Radeon RX 9070 XTPerfect · 2.6× 16 GB · $649–$779
- Mac Mini M4 16 GBPerfect · 1.7× 16 GB unified · $799 (new floor) / $499–$599 (eBay/residuals)
- AMD Radeon RX 7900 XTXPerfect · 3.9× 24 GB · $810 used / ~$1,340 new
- NVIDIA RTX 3090 (used, single)Perfect · 3.9× 24 GB · $950–$1,200
- NVIDIA RTX 5070 TiPerfect · 2.6× 16 GB · $980–$1,300
- NVIDIA RTX 5080Perfect · 2.6× 16 GB · $999–$1,400
- MacBook Air M5 24 GBPerfect · 2.6× 24 GB unified · $1,499–$1,899
- Mac Mini M4 Pro 24 GBPerfect · 2.6× 24 GB unified · $1,599
- Dual RTX 3090 (used)Perfect · 7.8× 48 GB · $1,800–$2,500 all-in
- Framework Desktop (Ryzen AI Max+ 395)Perfect · 13.9× 128 GB unified · $1,999–$2,851
- NVIDIA RTX 4090Perfect · 3.9× 24 GB · $2,200–$2,800
- M5 Pro MacBook Pro 48 GBPerfect · 5.2× 48 GB unified · $2,999–$3,599
- NVIDIA RTX 5090Perfect · 5.2× 32 GB · $3,500–$4,300
- NVIDIA RTX A6000 (48 GB, used)Perfect · 7.8× 48 GB ECC · $3,500–$4,500
- Mac Studio M4 Max 64 GBPerfect · 6.9× 64 GB unified · $3,799
- NVIDIA DGX SparkPerfect · 13.9× 128 GB unified · $4,699
- M5 Max MacBook Pro 64 GBPerfect · 6.9× 64 GB unified · ~$5,199 (est.; June 25 2026 increase)
- Mac Studio M3 Ultra 96 GBPerfect · 10.4× 96 GB unified · $5,299
- Dual RTX 5090Perfect · 10.3× 64 GB (2×32) · $8,500–$10,500
Next step
Find-by-model — see what hardware runs this→