Model · stat sheet
Yi-2 Large
01.AI's flagship open model; strong general reasoning with a large context window.
What it is
Yi-2 Large is 01.AI's flagship open model — the note on this page pairs strong general reasoning with a large context window. 01.AI, the Chinese lab founded by Kai-Fu Lee, built the original Yi line as open weights aimed at both Chinese and English use, and the large-tier model is where the general capability sits.
One caveat to carry into any decision: 01.AI has been through visible changes since the first Yi releases, and the status of the project and its support should be verified on the lab's channels before you build on the weights. This page measures the model as scored on 15 Sep 2026.
How the composite is built
Overall 78 weights Coding 24%, Terminal 20%, Reasoning 20%, Tool use 15%, Context 13%, Speed 8%. Yi-2 Large runs 74 to 82 across the six axes — a mid-band profile with Context the best of it and Terminal the weakest.
- Coding 80
- Level with Gemma 3 27B and GPT-5.6 Luna, and twelve behind DeepSeek V4's 92. Ordinary code work with review.
- Terminal 74
- Level with Mistral Small 3.2 and Gemma 3 27B. The weakest axis; keep shell work short and checked.
- Reasoning 81
- Just below Gemma 3 27B's 82 and two above Grok 5 Mini's 79. The note's "strong general reasoning" claim holds within its class, not against the 85-plus group.
- Tool use 76
- Two behind Gemma 3 27B's 78 and nine behind Haiku 4.5's 85. Expect error accumulation in long tool chains.
- Context 82
- Level with Llama 5 Scout and Grok 5 Mini — mid-field, despite the note's "large context window" framing. Verify the actual window in the model card rather than inferring it from the label.
- Speed 76
- Level with Qwen 4 Max's 76 and behind the efficient small tiers by a wide margin. Self-hosted throughput is a property of your hardware.
Where it fits
General work on a budget where open weights are required and the Qwen or Llama ecosystems are not a fit: bilingual Chinese–English products, research deployments, fine-tuning experiments. Its scores put it in the middle of the open-weights pack, so the case for choosing it over the neighbours has to come from language coverage, licence or supply-chain preference rather than from this board.
Limits
- No price or index is published here, so Value is Unrated and no cost comparison with the priced models is possible from this page.
- Terminal 74 and Tool use 76 are the practical constraints. This is not an agentic model.
- Context 82 is mid-field, so treat the "large window" framing on this page as something to check against the model card, not as a verified fact.
- Project status needs checking. 01.AI has been through restructuring since the original Yi releases; licence and support terms can change with it.
Price and access
No published price or gateway listing on this board. Access is the weights from 01.AI's channels under the published licence, run on your own hardware or a third-party host. Last scored 15 Sep 2026.
Alternatives on this board
- Gemma 3 27B — 79 at $0.08/$0.45 with open weights and a nearly identical profile at a published rate.
- Mistral Small 3.2 — 79 at $0.09/$0.25 with Speed 88, the cheaper hosted open option.
- Qwen 4 Max — 85 with Reasoning 87 and Context 87, the stronger multilingual flagship.
- Llama 5 Scout — 81 with Speed 85 and a larger ecosystem behind it.
Sources
- The 01.AI site for project announcements and current status.
- The 01.AI organisation on Hugging Face — weights, model cards and licence terms.
- On this site: local LLM inference, September 2026, and the full model board.
Scores are fullauto.online's composite index (0–100): Coding 24% · Terminal 20% · Reasoning 20% · Tool use 15% · Context 13% · Speed 8%. Editorial, not a vendor benchmark; 2026 tiers are early reads. Last scored 15 Sep 2026 · back to the leaderboard.