Ornith 1.0 35B

Ornith-1.0-35B is a text-generation model optimized for agentic coding tasks, built on top of pretrained Gemma 4 and Qwen 3.5 architectures using a self-improving reinforcement learning framework.

Self-Improving Training Framework

The model employs a two-stage RL process: it first proposes a refined scaffold conditioned on the task and previously used scaffold, then generates a solution rollout conditioned on that scaffold and task description. Reward from the rollout is propagated to both stages, enabling the model to discover better search trajectories and generate higher-quality solutions. A three-layer reward hacking defense protects training integrity through an immutable outer trust boundary, deterministic monitoring of forbidden actions, and a frozen LLM judge for intent-level gaming detection.

Benchmark Performance

Ornith-1.0-35B achieves state-of-the-art results among open-source models of comparable size on agentic coding benchmarks:

Benchmark	Ornith-1.0-35B	Qwen3.5-35B	Qwen3.6-35B	Gemma4-31B	Qwen3.5-397B
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

The 35B model surpasses Qwen 3.5-397B (53.5) on Terminal-Bench 2.1 despite having roughly one-tenth the parameters. The larger Ornith-1.0-397B variant achieves 77.5 on Terminal-Bench 2.1 and 82.4 on SWE-Bench Verified.

The model uses a structured tool-calling format with <tool_call> XML tags and nested <function=...> blocks. It is a reasoning model that opens assistant turns with a <think>...</think> block. Released under the MIT license.