Model components / Base / pretrained models

Yi-1.5

Kai-Fu Lee's 01.AI foundation model family in 6B, 9B, and 34B variants, with continued pretraining on 500B tokens over Yi-1.0. Strong performance on Chinese and English benchmarks. Apache 2.0 licensed. One of the top Chinese-origin open-weights models, though increasingly competing with Qwen for mindshare in the Chinese open source community.

Yi-1.5 base family (6B/9B/34B + 32K context variants), open sourced 2024-05-13 by 01.AI. Continuously pre-trained on Yi with +500B tokens (3.6T total). Base SKUs (Yi-1.5-9B, Yi-1.5-34B) scored here; Yi-1.5-*-Chat are the instruct SKUs (finetuned_chat). Verified live on HF June 2026; a dated 2024-era generation vs 2026 frontier.

Openness

3 high confidence

3.0

weights: open(Apache-2.0, on HF)
data: closed(pretraining corpus not released)
code: partial(inference/usage examples
license: Apache-2.0(OSI, permissive, no use restrictions)

Permissive Apache-2.0 weights but training data and full training code are not released, so open_weights not open_source. Apache-2.0 confirmed on HF base cards 2026-06-04.

https://huggingface.co/01-ai/Yi-1.5-34B recorded 2026-06-04
Apache-2.0 license, 34B base model (base vs chat variants listed), downloadable weights
https://huggingface.co/01-ai/Yi-1.5-9B recorded 2026-06-04
Apache-2.0 license, 9B base model, base/chat variants listed

Adoption

2 medium confidence

2.0

HF monthly downloads are modest in 2026 (Yi-1.5-34B 10,925/mo, Yi-1.5-9B 16,052/mo, both confirmed 2026-06-04) as a 2024-era family superseded by Qwen/Llama/DeepSeek generations. Cumulative reach across variants + ModelScope/wisemodel mirrors lands in the early-adopter band; relevance has faded.

https://huggingface.co/01-ai/Yi-1.5-34B recorded 2026-06-04
10,925 downloads last month
https://huggingface.co/01-ai/Yi-1.5-9B recorded 2026-06-04
16,052 downloads last month

Capability

2 medium confidence

2.0

Strong for a 2024 sub-35B model but well below 2026 frontier open models (DeepSeek-V4, Qwen 3.6); calibrates near Falcon 3 (C2) on the dated-generation ladder. Benchmarks from the 01-ai GitHub repo (not the HF card).

https://arxiv.org/abs/2403.04652 recorded 2026-06-04
Yi foundation-models tech report with benchmark methodology
https://github.com/01-ai/Yi-1.5 recorded 2026-06-04
Yi-1.5 base benchmark results (MMLU 76.3, BBH 54.3, C-Eval 81.4)

Unchanged since 2026-07-30 (last edited, not re-checked)