Yi (01.AI)

tl;dr: Yi 1.5 is 01.AI’s open bilingual family — three dense sizes (6B/9B/34B), trained on 3 trillion tokens of mixed English + Chinese. The Ollama listing hasn’t been refreshed in two years, so what you pull is a 2024 snapshot.

What it is

yi is the umbrella tag Ollama ships for 01.AI’s Yi 1.5 release. The library page’s own summary is one sentence: “Yi 1.5 is a high-performing, bilingual language model.” The README on the page is also one sentence: trained on a high-quality corpus of 3T tokens supporting English and Chinese.

The library lists 174 model variants (mostly quantizations of the same three sizes). The default yi lands on the 6B.

Sizes and hardware

All variants are text-only. No vision, no audio, no tool-calling tags on the page.

Pull

ollama run yi            # defaults to yi:6b
ollama run yi:9b
ollama run yi:34b

License

The ollama.com page does not state a license here. Upstream Yi is published under the Apache 2.0 license at github.com/01-ai/Yi and huggingface.co/01-ai/Yi-34B; check the upstream repo before commercial deployment.

What the README actually says

Almost nothing. Two sentences and a HuggingFace reference link. No benchmark numbers, no changelog, no per-size notes. The “Updated 2 years ago” badge on the listing matches what the page looks like — 2024’s snapshot.

Gotchas

• Last update was 2 years ago. For current Chinese-language models, look at Qwen3, GLM-5, or DeepSeek before defaulting to Yi.
• 4K context. Long-document use cases need a different model.
• The 34B fits in 19GB on disk, which means Q4_0 quantization. If you need higher fidelity, pull a non-default tag from the /library/yi/tags table.
• License is upstream-only — verify at github.com/01-ai/Yi.

Related: Qwen3 for the current Chinese/English open-weight default, DeepSeek-R1 for current reasoning distillates from the same era, Ollama for the runtime.

Untested here — facts from ollama.com/library/yi and the upstream Yi repo fetched 2026-09-10.