01.AI Yi (零一万物)
Listed at https://platform.lingyiwanwu.com
💰 Token Pricing
| Type | Price | Note |
|---|---|---|
| Input | ¥0.99/1M total tokens（Yi-Lightning，input+output 合并计费） | per million tokens |
| Output | ¥0.99/1M total tokens（合并计费，不区分 input/output） | per million tokens |
🤖 Supported Models (2)
✨ Pros
- ✓Founded by Dr. Kai-Fu Lee, globally recognized AI lab
- ✓Yi-Lightning at ¥0.99/1M tokens — extremely competitive pricing
- ✓Smart routing: auto-selects DeepSeek-V3 / Qwen3.5 / Yi-Lightning per request
- ✓100% OpenAI-compatible API /v1/chat/completions, zero SDK migration
- ✓Strong Chinese model ecosystem with balanced bilingual performance
- ✓Open-source model series (Yi-1.5 / Yi-VL / Yi-Coder) with broad community impact
⚠️ Cons
- ×Limited API model selection: only 2 models (Yi-Lightning + Yi-Vision-v2), no direct Llama/Qwen access
- ×Combined billing is opaque — cannot separately account for input vs output costs
- ×English documentation and community support weaker than international vendors
- ×No free initial credits — must recharge to use
- ×Smart routing model selection is non-deterministic (users cannot pin to DeepSeek-V3)
🎯 Best For
Cost-sensitive inference workloads in China; budget-conscious bilingual chat; teams migrating from OpenAI to low-cost domestic alternatives; smart-routing scenarios (no fixed model needed)
💰 Pricing & Plans
| Model | Route Target | Billing | Rate (¥/M total tokens) | Notes |
|---|---|---|---|---|
| Yi-Lightning | Yi-Lightning (smart-routed) | Combined (input + output) | ¥0.99 | Flagship chat, ultra-low price |
| Yi-Lightning | → DeepSeek-V3 (smart-routed) | Combined (input + output) | ¥0.99 | Same price regardless of which underlying model wins the route |
| Yi-Lightning | → Qwen3-30B-A3B (smart-routed) | Combined (input + output) | ¥0.99 | Smart router picks one of Lightning / DeepSeek / Qwen per request |
| Yi-Vision-v2 | Qwen2.5-VL-72B (smart-routed) | Combined (input + output) | ¥6.00 | Vision tier — image+text input |
| Yi-Vision-v2 | → Yi-Vision-V2 (smart-routed) | Combined (input + output) | ¥6.00 | Native Yi vision model behind the same endpoint |
🔧 API & Developer Experience
- •API Style: OpenAI-compatible REST endpoint at https://api.lingyiwanwu.com/v1/chat/completions — drop-in for the official OpenAI Python/Node SDKs by swapping base_url and api_key.
- •Smart Routing: Yi-Lightning is a meta-endpoint: each request is routed by the platform to Yi-Lightning / DeepSeek-V3 / Qwen3-30B-A3B based on latency/cost heuristics. Users do not pin which model wins (this is the documented non-deterministic behavior).
- •Vision Endpoint: Yi-Vision-v2 accepts image inputs through the same OpenAI-style messages API (image_url content parts). Behind the scenes it routes between Qwen2.5-VL-72B and Yi-Vision-V2.
- •SDK / Streaming: Works with the OpenAI SDK out of the box; SSE streaming supported on both endpoints. No first-party Python SDK beyond the OpenAI-compatible client.
- •Auth & Billing: API keys issued from the platform.lingyiwanwu.com console; prepaid CNY recharge required (no monthly invoicing for self-serve). Free tier is limited to a low RPM/TPM rate limit on registration — no initial credit balance.
- •Context Length: Yi-Lightning: 16K context window (per public docs). Yi-Vision-v2: ~8K token image+text budget. Long-context workloads are not Yi-Lightning's strength.
- •Rate Limits: Default rate limits scale with recharge history — typical first-recharge accounts get 60 RPM / 60K TPM; enterprise tiers can be negotiated up. Free accounts hit the lowest RPM/TPM cap and see soft-throttled requests rather than hard 429s.
🧠 Smart Routing: One Endpoint, Three Models
01.AI's defining capability is the smart-routing layer behind Yi-Lightning: a single endpoint exposes Yi-Lightning, DeepSeek-V3, and Qwen3-30B-A3B as routing candidates, and the platform picks a model per request based on latency and cost heuristics. Teams migrating from OpenAI get to point a single base_url at api.lingyiwanwu.com and let the platform route — at ¥0.99 per million total tokens (input + output combined) regardless of which underlying model wins. The trade-off is that the routing decision is non-deterministic: users cannot pin to DeepSeek-V3 specifically, so workloads that require exact reproducibility of which model produced an answer need a different provider. Yi-Vision-v2 mirrors this pattern for image understanding, routing between Qwen2.5-VL-72B and Yi-Vision-V2 at ¥6 per million total tokens.
🌐 China Access & Latency
01.AI is a Beijing-based AI lab and the api.lingyiwanwu.com endpoint is native-first for mainland China: directly reachable without a proxy, no VPN required, and supports domestic registration via Chinese mobile number plus Alipay / WeChat Pay for recharge. Domestic latency is typically 30-80 ms RTT from Beijing / Shanghai / Guangzhou, and the platform publishes 99.9% monthly availability on its enterprise tier. For China-based production workloads, this is one of the most cost-effective low-latency options in the 2026 landscape — the ¥0.99/1M combined rate undercuts most domestic rivals for chat workloads, while the Yi-Vision-v2 endpoint covers vision at ¥6/1M. For overseas access, api.lingyiwanwu.com is reachable but uncurated for non-China traffic; latency from North America is typically 250-450 ms and the platform does not advertise an SLA for international regions. Most overseas users who want Yi-Lightning route through Chinese aggregators (ModelScope, OpenRouter) or deploy DeepSeek-V3 / Qwen directly.