Sage-A1.5 vs Ornith-1.5-122B

Official Sage-A1.5 benchmark overview figure from the InternScience model card — InternScience/Sage-A1.5 card figures/Sage-A1p5-benchmark.png. Source · License: Apache-2.0.

Official Ornith-1.5-122B benchmark figure from the Hugging Face model card — Ornith-1.5-122B card assets/ornith_122b_eval.png. Source · License: MIT.

Verdict. Pick Sage-A1.5 if you want Shanghai AI Lab’s Apache-2.0 35B MoE VLM aimed at long-horizon science agents (orchestration + opt-in verification, 256K native). Pick Ornith-1.5-122B if you want a MIT 122B text coding/agent MoE the card says serves on one 80 GB GPU, with published vLLM/SGLang recipes. Sage’s card table includes an Ornith-1.5 column; those rows stay vendor. We have no sourced bake-off.

Compare Sage-A1.5 and Ornith-1.5-122B.

Dimension	Sage-A1.5	Ornith-1.5-122B	Source
License	Apache-2.0 (card tag; LICENSE blob 404)	MIT (card tag; LICENSE blob 404 on raw fetch)	Hugging Face cards (2026-10-23 / 2026-10-22)
Gating	No	No	Hugging Face API
Developer	Shanghai AI Laboratory (InternScience)	Ornith	Official cards
Parameters	35B MoE on the card; HF safetensors ~36.0B	122B MoE on the card; HF safetensors ~125.1B	Official cards / HF API
Modalities	VLM (text + image); HF tags vlm / vision	text-generation	Official cards / HF API
Context length	256K native; recipes 262,144; YaRN factor 2.0 ~512K	262,144 native; YaRN factor 4.0 ~1M (vendor)	Official cards
Serving named	vLLM and SGLang with qwen3 / qwen3_coder parsers	vLLM ≥ 0.19.1; SGLang ≥ 0.5.9; Transformers ≥ 5.8.1	Official cards
Stated job	Long-horizon science / agent orchestration + verification	Coding and agents; single-GPU 122B MoE	Official cards
Hugging Face likes	223 on the BF16 card (2026-10-23)	129 on the BF16 card (2026-10-22)	Hugging Face API
Quality vs ChatGPT	Not independently tested	Not independently tested	n/a

When to choose which

Multimodal (image + text in): Sage-A1.5 — Sage is tagged VLM. Ornith’s pipeline_tag is text-generation.
Published 122B MIT coding/agent card: Ornith-1.5-122B — Size and license identification. Not a SWE-bench ranking.
Opt-in orchestration / verification protocol: Sage-A1.5 — Card documents spawn_subagent and chat_template_kwargs.verification. Untested here.
Native context near 256–262K without YaRN: Tie — Sage README 256K; recipes 262,144. Ornith native 262,144.

Read the entity pages: Sage-A1.5 and Ornith-1.5-122B. Both hang off ChatGPT alternatives and closed LLM APIs. Ornith also hangs off Claude Code alternatives.

Why this comparison exists

Both families published ungated agent-oriented MoE weights in October 2026. Sage-A1.5 is the ~35B science/VLM card (created 2026-10-15; paper 2026-10-21). Ornith-1.5-122B is the 122B text coding card (created 2026-10-16). Sage’s own table already names Ornith-1.5; this page records identification facts so we do not treat that table as a forkalt ranking.

Sourced snapshot (2026-10-23 / Ornith 2026-10-22)

Field	Sage-A1.5	Ornith-1.5-122B
Host	huggingface.co/InternScience/Sage-A1.5 (not gated)	huggingface.co/ornith-ai/Ornith-1.5-122B (not gated)
HF likes / downloads	223 / 396 (2026-10-23)	129 / 742 (2026-10-22)
License	Apache-2.0 (card tag; LICENSE 404)	MIT (card tag; LICENSE 404)
Size	35B MoE (card); ~36.0B safetensors	122B MoE (card); ~125.1B safetensors
Job	VLM; science / long-horizon agents	text-generation; coding/agents
Context	256K native	262,144 native
Official serve path	vLLM / SGLang + qwen3 parsers	vLLM / SGLang + qwen3 parsers
Disk (BF16 usedStorage)	~67.0 GiB	~233 GiB
Quality vs ChatGPT	Not independently tested	Not independently tested

License

Sage tags Apache-2.0. Ornith tags MIT. Both LICENSE blobs 404’d as raw files in the checks we opened. Not legal advice.

How to choose

Need Apache-2.0 35B science/agent VLM with orchestration/verification flags: Sage-A1.5.
Need MIT 122B coding/agent weights and a single-80GB card claim: Ornith-1.5-122B.
Need Apache-2.0 40B-A2.5B local coding with Ollama/GGUF: Flint — Flint vs Ornith-1.5-122B.
Need Ornith versus DeepSeek’s 1017 checkpoint: Ornith-1.5-122B vs DeepSeek-V4.1-Flash.

FAQ

Does Sage’s card already compare against Ornith?

Yes. The Sage-A1.5 README table includes an Ornith-1.5 column on several benches. Those figures are vendor/self-reported. We do not copy them into a ranking.

Can I serve both with vLLM?

Both cards publish vLLM recipes with Qwen3 reasoning/tool-call parsers. Sage also names a --language-model-only flag. Untested here. See vLLM.

Which is smaller on disk?

Sage BF16 usedStorage is ~67.0 GiB. Ornith BF16 usedStorage is ~233 GiB. Those are Hugging Face disk figures, not peak VRAM.

Does either replace ChatGPT?

No sourced bake-off on this site. Both hang off those hubs as identification rows.

Sources & verification

Last verified 2026-10-23 · Last updated 2026-10-23· Edited by forkalt editors · Editorial policy

License notes are not legal advice. Read the original LICENSE. Trademark & disclaimer.