Ornith-1.5-9B

Ornith-1.5-9B

Quick answer: Ornith-1.5-9B is the most lightweight member of Ornith's (DeepReinforce's) Ornith-1.5 family — a 9B dense model designed for efficient single-GPU deployment and edge/mobile deployment via its quantized Ornith-1.5-9B-Mobile variant. Released under MIT license, it scores 70.6% on SWE-bench Verified and 46.2–47% on Terminal Bench 2.1, matching or exceeding much larger models like Gemma 4-31B and Qwen 3.6-35B.

At a Glance

Where Ornith-1.5-9B leads

• 70.6% SWE-bench Verified — matches or exceeds larger models like Gemma 4-31B (52.0%) despite being ~1/3 to ~1/4 the size
• 86.4% GPQA Diamond — strong graduate-level science reasoning for a 9B model
• 66.5% ClawEval, 59.5% WideSearch — solid agentic tool-use and search performance at edge scale
• MIT license — fully open weights, permissive for commercial use
• Edge/mobile-deployable via the quantized Ornith-1.5-9B-Mobile variant

Where it lags

• 47.5% SWE-bench Pro — notably behind SWE-bench Verified, following the pattern seen across all Ornith-1.5 sizes and most models generally
• 30.5% HLE (with tools) / 20.2% (no tools) — modest on frontier-exam reasoning, as expected at this parameter scale
• No DeepSWE or Frontier-Bench v0.1 scores reported on the model card at this size

Best for: single-GPU or edge/mobile self-hosted coding assistants and lightweight agentic tool-use, where a compact dense model needs to substantially outperform its size class.

What Ornith-1.5-9B Is

Ornith-1.5-9B is the most lightweight member of Ornith's (built by the DeepReinforce team) Ornith-1.5 model family, a 9B dense model designed for efficient single-GPU deployment (~19GB in BF16, fits on a single 80GB GPU) and edge-deployable on mobile devices via its quantized Ornith-1.5-9B-Mobile variant. Like its larger 35B-A3B and 397B siblings, it's trained via Ornith-1.5's self-improvement loop — jointly generating training tasks, task-specific scaffolds, and solution rollouts, optimized end-to-end with reinforcement learning.

Despite its compact size, Ornith-1.5-9B "delivers remarkably strong results," per the official model card, achieving 47.0% on Terminal Bench 2.1 (Claude Code harness) and 70.6% on SWE-bench Verified — matching or exceeding much larger models such as Gemma 4-31B and Qwen 3.6-35B. It supports a 262,144-token context window, extensible to roughly 1,000,000 tokens via YaRN RoPE scaling.

Specs

Pricing

Ornith-1.5-9B is released as open weights (self-hosted); Ornith/DeepReinforce has not announced a hosted API pricing tier for this model.

Public Benchmark Scores

Scores are self-reported by the Ornith team on the official Hugging Face model card (ornith-ai/Ornith-1.5-9B) and the accompanying technical blog post; the top 4 rows are additionally confirmed via the card's structured HF "Evaluation Results" metadata. All results are averaged over 5 independent runs. HLE score uses the "with tools" evaluation condition (20.2% without tools). Note: DeepSWE and Frontier-Bench v0.1 are not reported for this model size on the official benchmark table.

Notable Results Not Yet Tracked as Benchgen Benchmarks

• SWE-bench Multilingual (54.4%), NL2Repo (32.4%), SWE Atlas – QnA (20.6%), and WideSearch (59.5%) are all real, methodology-documented benchmarks per the model card's footnotes, but don't have existing Benchgen benchmark pages yet — flagged as follow-ups (subset of the set flagged on the sibling Ornith-1.5-397B and 35B-A3B pages; this smaller model wasn't evaluated on Frontier-Bench v0.1 or ClawEval-adjacent DeepSWE per the official table, though ClawEval itself is reported at 66.5%).

Ornith-1.5-9B vs Alternatives

¹ Comparative figures per the Ornith-1.5 model card's cross-model benchmark table; verify directly against each model's own page for precise sourcing.

FAQ

Is Ornith-1.5-9B open source? Yes — MIT license, with weights available on Hugging Face (ornith-ai/Ornith-1.5-9B), plus GGUF and MLX quantized variants.

Who makes Ornith-1.5-9B? Ornith, built by the DeepReinforce team.

How big is Ornith-1.5-9B? 9 billion parameters, dense (not mixture-of-experts).

What is Ornith-1.5-9B's context window? 262,144 tokens natively, extensible to roughly 1,000,000 tokens via YaRN RoPE scaling.

Can Ornith-1.5-9B run on mobile devices? Yes — a quantized Ornith-1.5-9B-Mobile variant is designed for edge deployment on iPhone and Android devices.

Where can I access Ornith-1.5-9B? Self-hosted from the open weights at huggingface.co/ornith-ai/Ornith-1.5-9B, which fits on a single 80GB GPU in BF16 (~19GB).

Benchmark scores sourced from the official Hugging Face model card for ornith-ai/Ornith-1.5-9B and the Ornith team's technical blog (Aug 2026), cross-checked against the card's structured HF Eval Results metadata where available.