01.AI’s open-weight LLM family, developed in Beijing by Kai-Fu Lee’s lab starting March 2023. 01.AI achieved unicorn status within eight months and released Yi-34B, Yi-Lightning, and Yi-Coder across 2023 and 2024. Yi-Lightning briefly ranked joint third on Chatbot Arena at its October 2024 launch.
Status as of June 2026: maintenance Yi, headline WorldWise/WanZhi. platform and the Chinese WanZhi site, now at version 2.5. The public GitHub news stream still centers 2024 Yi releases, while the 2026 company site highlights multi-agent enterprise deployment and Super Employee agent products. Yi-2 is not announced.
System Verdict
Pick Yi for narrow legacy use cases or as the on-ramp to WorldWise/WanZhi. Yi-34B-200K still works as a fixed baseline for research. Yi-Coder and Yi-1.5 remain available through public model paths for teams that have already integrated them. Yi-Lightning’s official direct pricing is no longer surfaced on the June 2026 pages AiPedia checked, so treat the old ~$0.14/M-token figure as historical unless a gateway quote confirms it. Enterprises that want 01.AI’s WorldWise/WanZhi 2.5 multi-agent platform can evaluate that directly.
Skip it for new frontier production. The family receives no major pre-training updates. DeepSeek delivers what Yi-Lightning promised with an active roadmap. Qwen offers a broader open-weight family under Apache 2.0 with monthly releases. Kimi K2.6 lands the open-weight agentic baseline that 01.AI has stepped back from.
Who still uses Yi: Academic researchers freezing a fixed baseline, hobbyists running Yi-34B-200K on existing infrastructure, teams with pre-existing Yi-Coder deployments, and enterprises buying WorldWise 2.5 from 01.AI’s go-to-market team.
Key Facts
| Status | Yi family in maintenance since the March 2025 pre-training pause |
| Current focus of 01.AI | WorldWise/WanZhi enterprise LLM platform, version 2.5 in 2026 |
| Highest-performing Yi model | Yi-Lightning (October 2024) |
| Flagship open-weight | Yi-34B-200K (4K base, 200K extended context) |
| Coding models | Yi-Coder-9B and Yi-Coder-1.5B public model paths |
| Enterprise platform | WorldWise/WanZhi 2.5 multi-agent system, “Super Employee” agent products, CEO-led go-to-market |
| Yi-Lightning pricing | Historical ~$0.14/M input and output; no current direct 01.AI price surfaced in June 2026 source check |
| Third-party access | OpenRouter, AIMLAPI, other aggregators |
| License posture | GitHub now shows Apache-2.0 license files for Yi-1.5; 01.AI site still routes commercial-license applications, so confirm model-specific terms |
| No Yi-2 announced | Public 01.AI and GitHub pages do not show a new Yi-2 release |
Every data point above was verified on 2026-06-12. See Sources.
What it actually is
family paired with an enterprise platform. The Yi weights remain discoverable through 01.AI-linked Hugging Face and GitHub paths. Third-party API enterprise platform.
The original Yi product delivered strong results. Yi-34B beat Llama 2 at launch. Yi-Lightning ranked competitively on Chatbot Arena at $0.14/M tokens, one of the best price-per-quality ratios of its era. Yi-Coder is a capable open coding model under Apache 2.0.
The strategic story has shifted twice. In March 2025 Kai-Fu Lee publicly argued that pre-training below hundred-million-user scale is uneconomic, and 01.AI moved consulting workloads onto DeepSeek-backed agent stacks. In 2026 the company packaged that pivot into WorldWise/WanZhi 2.5, a multi-agent enterprise platform sold directly by 01.AI’s go-to-market organization. Yi sits beside that platform as a legacy artifact and an open-weight calling card.
When to pick Yi
- Fixed-weight research baselines. Academic papers benefit from models that will not change. Yi-34B-200K gives a citable frozen target.
- Long-context experiments on older hardware. Yi-34B-200K runs on a single A100 and handles 200K tokens for research prototypes.
- Legacy Yi-Coder deployments. Teams already using Yi-Coder-9B or Yi-Coder-1.5B can continue running those models, but should confirm model-specific license text before commercial redistribution.
- Bilingual Chinese-English tasks with fixed scope. Yi’s training still handles both languages competently for narrow-scope workloads.
- WorldWise/WanZhi 2.5 enterprise evaluation. Buyers exploring 01.AI’s multi-agent enterprise platform get Yi positioning as part of the package.
- Budget API through third-party gateways. Yi-Lightning may remain available through aggregators, but price and SLA need provider-level confirmation.
When to pick something else
- Active open-weight development: Qwen releases monthly under Apache 2.0 across 119 languages.
- Cheapest capable API with a roadmap: DeepSeek delivers Yi-Lightning’s value proposition with ongoing model updates.
- Polished English product: ChatGPT or Claude.
- Open-weight coding leader: GLM GLM-5.1 under MIT at SWE-Bench Pro leader tier.
- Open-weight agentic baseline: Kimi K2.6 Agent Swarm leads the open-weight agentic category that 01.AI has stepped away from.
- Long-context frontier model: Claude Opus 4.8 at 1M tokens, or Kimi K2.6 at 256K.
Pricing
Yi-Lightning pricing depends on third-party aggregators unless 01.AI sales confirms a direct route. Yi weights remain linked through Hugging Face 01-ai and GitHub. WorldWise/WanZhi enterprise pricing is quoted directly by 01.AI.
| Surface | Price | Notes |
|---|---|---|
| Yi open weights | Free to download where hosted | Confirm model-specific license; GitHub lists Apache-2.0 files for Yi-1.5 while 01.AI still routes commercial-license applications |
| Yi-Coder-9B / 1.5B | Free where hosted | Confirm model-specific license before commercial redistribution |
| Yi-Lightning (direct, historical) | ~$0.14/M input and output | Historical October 2024 figure; current direct 01.AI pricing not surfaced in June 2026 official pages |
| Yi-Lightning (third-party gateways) | Varies | OpenRouter, AIMLAPI, other aggregators |
| WorldWise/WanZhi 2.5 enterprise platform | Custom | Multi-agent platform; CEO-led sales motion in 2026 |
Prices verified 2026-06-12 via 01.AI company site, 01.AI Yi models, Hugging Face 01-ai, Yi GitHub, and SCMP coverage of 01.AI’s enterprise pivot. Direct 01.AI API access post-pivot remains uncertain; third-party gateways are the practical path for Yi-Lightning unless sales confirms otherwise, while WorldWise/WanZhi 2.5 is a sales-led enterprise platform.
Against the alternatives
| Yi (frozen) | DeepSeek V3 | Qwen3 open-weight | GLM-5.1 | |
|---|---|---|---|---|
| Active development | No (WorldWise platform active) | Yes | Yes (monthly) | Yes |
| Open weights | Model-specific; verify current license files | V3 open | Apache 2.0 | MIT |
| Current benchmarks | Frozen at 2024 level | Frontier reasoning | Frontier multilingual | SWE-Bench Pro leader |
| API price | ~$0.14/M (third-party) | ~$0.28/M | ~$0.15-$0.325/M | $1.00/M |
| Context window | 200K (Yi-34B-200K) | 64K | 1M (Qwen3.6 Plus) | 200K |
| Vendor roadmap | None for new Yi; WorldWise 2.5 active | Active | Active | Active |
| Best viewed as | Legacy research baseline plus WorldWise on-ramp | Cheap capable API | Open-weight multilingual | Open-weight coding leader |
Failure modes
- No visible Yi-2 roadmap. The June 2026 official pages checked by AiPedia still position WorldWise/WanZhi rather than a new Yi generation, and GitHub’s visible news remains 2024-centered.
- Yi-Lightning API availability is uncertain. Direct 01.AI hosted access post-pivot is ambiguous. Third-party gateways remain the practical path, but per-provider SLAs vary.
- License ambiguity by model. GitHub shows Apache-2.0 files for Yi-1.5, while 01.AI’s official model page still routes commercial-license applications. Treat commercial deployment as model-specific legal review, not a blanket green light.
- Community support is less active than current frontier alternatives. The public repo remains useful, but it does not show the same 2026 release cadence as Qwen, DeepSeek, Kimi, or GLM.
- Enterprise path flows through WorldWise/WanZhi. 01.AI’s WorldWise/WanZhi 2.5 platform is the company’s go-to-market headline. Customers buying from 01.AI are mostly buying the agent platform, with Yi as a brand asset rather than the underlying frontier model.
- Benchmark drift. Yi-Lightning’s late-2024 Chatbot Arena ranking does not reflect 2026 frontier scores. DeepSeek V3, Qwen3, GLM-5.1, and Kimi K2.6 have all surpassed it.
- Same Chinese-provider data residency posture as active vendors. Without the benefit of an active Yi roadmap to justify the tradeoff.
Methodology
This page was produced by the aipedia.wiki editorial pipeline, an automated system that ingests vendor documentation, verifies pricing and model details against primary sources, and generates the editorial analysis you are reading. No individual human wrote this review. Scoring follows the four-dimension rubric at /about/scoring/ (Utility, Value, Moat, Longevity; unweighted average). Longevity is scored low to reflect the frozen-model status of the Yi family, even though 01.AI as a company remains operational via WorldWise/WanZhi 2.5. Last verified 2026-06-12 against 01.AI company site, 01.AI Yi models, Lingyiwanwu/WanZhi site, SCMP interview with Kai-Fu Lee, Hugging Face 01-ai, and Yi GitHub.
FAQ
Is 01.AI still developing Yi? The public June 2026 pages AiPedia checked keep Yi-Lightning and Yi open models listed, but 01.AI’s active product energy goes into WorldWise/WanZhi 2.5, an enterprise multi-agent platform. No Yi-2 is announced on the official site or GitHub page.
Is Yi open source? Mixed enough to require model-specific review. GitHub now shows Apache-2.0 license files and says Yi-1.5 code and weights use Apache 2.0, while 01.AI’s official model page still routes commercial-license applications. Check the exact checkpoint license before commercial deployment.
How does Yi-Lightning compare in 2026? At its October 2024 launch, Yi-Lightning ranked joint third on Chatbot Arena at a historical ~$0.14/M-token price. Since then, active model families such as DeepSeek, Qwen, GLM, and Kimi have moved faster. 01.AI still labels Yi-Lightning “Global SOTA MoE era.
Should I start a new project on Yi? Only if you specifically need a frozen-weight research baseline or already have Yi-Coder integrated. For new production work, DeepSeek, Qwen, GLM, and Kimi offer active roadmaps at similar or better price points.
What is WorldWise? 01.AI’s enterprise LLM platform, now at version 2.5. The English site calls it WorldWise; the Chinese site brands the platform as WanZhi. It packages multi-agent “Super Employee” deployments for business customers and is the headline product on 01.AI’s 2026 corporate site.
Sources
- 01.AI company site: WorldWise 2.5 positioning and Yi-Lightning marketing, verified 2026-06-12
- 01.AI Yi models: Yi-Lightning, Yi-1.5, Yi-VL, and Yi-Coder public model positioning, verified 2026-06-12
- Lingyiwanwu/WanZhi site: Chinese WanZhi enterprise-agent positioning and January 5, 2026 WanZhi 2.5 item, verified 2026-06-12
- SCMP: Kai-Fu Lee bets on DeepSeek: strategic reversal interview, verified 2026-06-12
- Hugging Face 01-ai: current public model availability, verified 2026-06-12
- Yi open-source GitHub: repo license, release-news, and community status, verified 2026-06-12
- Yi: Open Foundation Models paper: original model architecture and training, verified 2026-06-12
Related
- Category: AI Research
Review History
- 2026-06-10: Re-verified 01.AI’s WorldWise/WanZhi 2.5 enterprise-agent positioning, Yi model page, Hugging Face organization, GitHub license/release-news posture, and direct-pricing uncertainty.
