Task-led research & comparison
Choose a Chinese AI model by the constraint that matters.
Start with coding, context, latency measurement, video, or value. Each route separates current model facts from the decision you still need to test. Token rates are synchronized from the gateway and may follow providers' official China list prices where configured. Media rows show transparent USD service rates; any service margin is included in the displayed rate.
This is a decision hub, not a popularity or quality leaderboard. We publish only current catalog facts: context window, listed USD price, modality, and declared capabilities. For independent quality benchmarks, see Artificial Analysis.
Start with the decision
Five routes, one evidence standard.
The route cards state the evidence we have, when it is useful, and where it stops. They lead to the relevant model facts and a single UTM-tagged registration path.
Coding & agents
Start with models whose current descriptions or capabilities explicitly mention coding, tools, agents, engineering, planning, or structured work.
- Use it when: Use this route when tool use, code generation, or an agent loop is the bottleneck.
- Limit: Capability tags describe declared fit; they are not a measured coding-quality score.
Long context
Use the current context-window field to narrow candidates before evaluating retrieval quality, prompt structure, and total token cost.
- Use it when: Use this route for large repositories, long documents, and multi-step knowledge work.
- Limit: A larger context window does not prove better recall, reasoning, or latency.
Low latency
ChinaAPI does not publish a cross-region latency ranking yet. This route gives you a repeatable way to measure your own prompt, region, and streaming mode.
- Use it when: Use this route when interaction time or first-token time is a release gate.
- Limit: Do not infer latency from listed price, context, or an individual request.
Video generation
Compare current video modalities, capabilities, and transparent per-unit USD service prices before testing your own prompt and duration.
- Use it when: Use this route for story, motion, reference-image, and audio requirements.
- Limit: The catalog is not a visual-quality or task-success ranking.
Value
Use listed input price as a cost filter, then test accepted-result cost with your workload, retries, output length, and operating constraints.
- Use it when: Use this route when budget is a hard constraint and a direct price comparison is useful.
- Limit: The lowest listed input price is not a quality, latency, or total-cost claim.
Model facts by task
Shortlist with facts, then validate your workload.
Models can appear in more than one group. Within these compact tables, rows are ordered by larger context window first, then lower listed input price.
Coding & agents
LLMs whose descriptions or capabilities reference coding, agents, tools, engineering, planning, or office work. This is catalog evidence, not a measured quality ranking.
| Model | Vendor | Context | Input $/1M |
|---|---|---|---|
| deepseek-v4-flash | DeepSeek | 1M | $0.14 |
| mimo-v2.5 | Xiaomi | 1M | $0.14 |
| qwen3.6-flash | Alibaba | 1M | $0.1846 |
| MiniMax-M3 | MiniMax | 1M | $0.2877 |
| LongCat-2.0 | Meituan | 1M | $0.3 |
| qwen3.7-plus | Alibaba | 1M | $0.3077 |
| deepseek-v4-pro | DeepSeek | 1M | $0.435 |
| mimo-v2.5-pro | Xiaomi | 1M | $0.435 |
| qwen3.6-plus | Alibaba | 1M | $0.5 |
| glm-5.2 | Zhipu AI | 1M | $1.4 |
| qwen3.8-max | Alibaba | 1M | $1.99 |
| qwen3.7-max | Alibaba | 1M | $2.5 |
| kimi-k3 | Moonshot | 1M | $2.7397 |
| step-3.5-flash | StepFun | 256K | $0.1 |
| step-3.5-flash-2603 | StepFun | 256K | $0.1 |
| hy3 | Tencent | 256K | $0.1538 |
| step-3.7-flash | StepFun | 256K | $0.2 |
| doubao-seed-2-1-turbo-260628 | ByteDance | 256K | $0.4615 |
| doubao-seed-2-1-pro-260628 | ByteDance | 256K | $0.9231 |
| kimi-k2.6 | Moonshot | 256K | $0.95 |
| kimi-k2.7-code | Moonshot | 256K | $0.95 |
| kimi-k2.7-code-highspeed | Moonshot | 256K | $1.7808 |
| MiniMax-M2.7 | MiniMax | 200K | $0.2877 |
| MiniMax-M2.7-highspeed | MiniMax | 200K | $0.5753 |
| glm-5 | Zhipu AI | 200K | $1 |
| glm-5-turbo | Zhipu AI | 200K | $1.2 |
| glm-5.1 | Zhipu AI | 200K | $1.4 |
| stepaudio-2.5-chat | StepFun | $1.5 | |
| step-1o-audio | StepFun | $3.85 |
Best for Reasoning
LLMs tagged with Reasoning in the current data source.
| Model | Vendor | Context | Input $/1M |
|---|---|---|---|
| deepseek-v4-flash | DeepSeek | 1M | $0.14 |
| mimo-v2.5 | Xiaomi | 1M | $0.14 |
| qwen3.6-flash | Alibaba | 1M | $0.1846 |
| MiniMax-M3 | MiniMax | 1M | $0.2877 |
| LongCat-2.0 | Meituan | 1M | $0.3 |
| qwen3.7-plus | Alibaba | 1M | $0.3077 |
| deepseek-v4-pro | DeepSeek | 1M | $0.435 |
| mimo-v2.5-pro | Xiaomi | 1M | $0.435 |
| qwen3.6-plus | Alibaba | 1M | $0.5 |
| glm-5.2 | Zhipu AI | 1M | $1.4 |
| qwen3.8-max | Alibaba | 1M | $1.99 |
| qwen3.7-max | Alibaba | 1M | $2.5 |
| kimi-k3 | Moonshot | 1M | $2.7397 |
| step-3.5-flash | StepFun | 256K | $0.1 |
| step-3.5-flash-2603 | StepFun | 256K | $0.1 |
| hy3 | Tencent | 256K | $0.1538 |
| step-3.7-flash | StepFun | 256K | $0.2 |
| doubao-seed-2-1-turbo-260628 | ByteDance | 256K | $0.4615 |
| doubao-seed-2-1-pro-260628 | ByteDance | 256K | $0.9231 |
| kimi-k2.6 | Moonshot | 256K | $0.95 |
| kimi-k2.7-code | Moonshot | 256K | $0.95 |
| kimi-k2.7-code-highspeed | Moonshot | 256K | $1.7808 |
| MiniMax-M2.7 | MiniMax | 200K | $0.2877 |
| MiniMax-M2.7-highspeed | MiniMax | 200K | $0.5753 |
| glm-5 | Zhipu AI | 200K | $1 |
| glm-5-turbo | Zhipu AI | 200K | $1.2 |
| glm-5.1 | Zhipu AI | 200K | $1.4 |
| stepaudio-2.5-chat | StepFun | $1.5 |
Long context
All LLMs with a 1M context window, ordered by lower listed input price.
| Model | Vendor | Context | Input $/1M |
|---|---|---|---|
| deepseek-v4-flash | DeepSeek | 1M | $0.14 |
| mimo-v2.5 | Xiaomi | 1M | $0.14 |
| qwen3.6-flash | Alibaba | 1M | $0.1846 |
| MiniMax-M3 | MiniMax | 1M | $0.2877 |
| LongCat-2.0 | Meituan | 1M | $0.3 |
| qwen3.7-plus | Alibaba | 1M | $0.3077 |
| deepseek-v4-pro | DeepSeek | 1M | $0.435 |
| mimo-v2.5-pro | Xiaomi | 1M | $0.435 |
| qwen3.6-plus | Alibaba | 1M | $0.5 |
| glm-5.2 | Zhipu AI | 1M | $1.4 |
| qwen3.8-max | Alibaba | 1M | $1.99 |
| qwen3.7-max | Alibaba | 1M | $2.5 |
| kimi-k3 | Moonshot | 1M | $2.7397 |
Multimodal
LLMs with vision=true. Text-only models are not listed in this group.
| Model | Vendor | Context | Input $/1M |
|---|---|---|---|
| mimo-v2.5 | Xiaomi | 1M | $0.14 |
| qwen3.6-flash | Alibaba | 1M | $0.1846 |
| MiniMax-M3 | MiniMax | 1M | $0.2877 |
| qwen3.7-plus | Alibaba | 1M | $0.3077 |
| mimo-v2.5-pro | Xiaomi | 1M | $0.435 |
| qwen3.6-plus | Alibaba | 1M | $0.5 |
| qwen3.8-max | Alibaba | 1M | $1.99 |
| kimi-k3 | Moonshot | 1M | $2.7397 |
| step-3.7-flash | StepFun | 256K | $0.2 |
| doubao-seed-2-1-turbo-260628 | ByteDance | 256K | $0.4615 |
| doubao-seed-2-1-pro-260628 | ByteDance | 256K | $0.9231 |
| kimi-k2.6 | Moonshot | 256K | $0.95 |
| kimi-k2.7-code | Moonshot | 256K | $0.95 |
| kimi-k2.7-code-highspeed | Moonshot | 256K | $1.7808 |
| step-1o-turbo-vision | StepFun | 32K | $0.4 |
From shortlist to implementation
Choose a model, verify its setup path, then register.
Model cards narrow the catalog; the integration guides provide the next configuration step. Guide links are shown only for tools with a dated verification record, so a catalog capability tag is never presented as a compatibility guarantee.
Continue with a verified setup guide
Open the tool guide that matches your workflow, then start with a small test before routing production traffic.
Low-latency validation
Measure the workflow you will actually ship.
We do not publish a cross-region latency ranking because the current catalog does not contain a repeatable multi-region latency dataset. Keep the model, prompt, request shape, region, concurrency, and streaming mode fixed when you test.
A repeatable latency check
Record time to first token and total completion time across multiple runs. Compare only runs with the same prompt and output limit, then keep the raw measurements with the test date and region.
Limit: a single request, a provider specification, or a listed price cannot support a latency claim. Until a documented probe publishes sample size and region, this hub will not label any model as the lowest-latency option.
Longest context window
All LLMs ordered by context.
Sorted by context window descending, then listed input price ascending. Output pricing is shown for token-billed models.
| Rank | Model | Vendor | Context | Input $/1M | Output $/1M |
|---|---|---|---|---|---|
| 1 | deepseek-v4-flash | DeepSeek | 1M | $0.14 | $0.28 |
| 2 | mimo-v2.5 | Xiaomi | 1M | $0.14 | $0.28 |
| 3 | qwen3.6-flash | Alibaba | 1M | $0.1846 | $1.1077 |
| 4 | MiniMax-M3 | MiniMax | 1M | $0.2877 | $1.1507 |
| 5 | LongCat-2.0 | Meituan | 1M | $0.3 | $1.2 |
| 6 | qwen3.7-plus | Alibaba | 1M | $0.3077 | $1.2308 |
| 7 | deepseek-v4-pro | DeepSeek | 1M | $0.435 | $0.87 |
| 8 | mimo-v2.5-pro | Xiaomi | 1M | $0.435 | $0.87 |
| 9 | qwen3.6-plus | Alibaba | 1M | $0.5 | $3 |
| 10 | glm-5.2 | Zhipu AI | 1M | $1.4 | $4.4 |
| 11 | qwen3.8-max | Alibaba | 1M | $1.99 | $5.97 |
| 12 | qwen3.7-max | Alibaba | 1M | $2.5 | $7.5 |
| 13 | kimi-k3 | Moonshot | 1M | $2.7397 | $13.6986 |
| 14 | step-3.5-flash | StepFun | 256K | $0.1 | $0.3 |
| 15 | step-3.5-flash-2603 | StepFun | 256K | $0.1 | $0.3 |
| 16 | hy3 | Tencent | 256K | $0.1538 | $0.6154 |
| 17 | step-3.7-flash | StepFun | 256K | $0.2 | $1.15 |
| 18 | doubao-seed-2-1-turbo-260628 | ByteDance | 256K | $0.4615 | $2.3077 |
| 19 | doubao-seed-2-1-pro-260628 | ByteDance | 256K | $0.9231 | $4.6154 |
| 20 | kimi-k2.6 | Moonshot | 256K | $0.95 | $3.9462 |
| 21 | kimi-k2.7-code | Moonshot | 256K | $0.95 | $3.9462 |
| 22 | kimi-k2.7-code-highspeed | Moonshot | 256K | $1.7808 | $7.3973 |
| 23 | MiniMax-M2.7 | MiniMax | 200K | $0.2877 | $1.1507 |
| 24 | MiniMax-M2.7-highspeed | MiniMax | 200K | $0.5753 | $2.3014 |
| 25 | glm-5 | Zhipu AI | 200K | $1 | $3.6667 |
| 26 | glm-5-turbo | Zhipu AI | 200K | $1.2 | $4.4571 |
| 27 | glm-5.1 | Zhipu AI | 200K | $1.4 | $4.4 |
| 28 | step-1o-turbo-vision | StepFun | 32K | $0.4 | $1.28 |
| 29 | stepaudio-2.5-chat | StepFun | $1.5 | $3.5 | |
| 30 | step-1o-audio | StepFun | $3.85 | $9.24 |
Value filter
LLMs ordered by listed input price.
Sorted by listed input USD per 1M tokens, ascending. Use this as a cost filter, not a quality, latency, or accepted-result-cost ranking.
| Rank | Model | Vendor | Context | Input $/1M | Output $/1M |
|---|---|---|---|---|---|
| 1 | step-3.5-flash | StepFun | 256K | $0.1 | $0.3 |
| 2 | step-3.5-flash-2603 | StepFun | 256K | $0.1 | $0.3 |
| 3 | deepseek-v4-flash | DeepSeek | 1M | $0.14 | $0.28 |
| 4 | mimo-v2.5 | Xiaomi | 1M | $0.14 | $0.28 |
| 5 | hy3 | Tencent | 256K | $0.1538 | $0.6154 |
| 6 | qwen3.6-flash | Alibaba | 1M | $0.1846 | $1.1077 |
| 7 | step-3.7-flash | StepFun | 256K | $0.2 | $1.15 |
| 8 | MiniMax-M3 | MiniMax | 1M | $0.2877 | $1.1507 |
| 9 | MiniMax-M2.7 | MiniMax | 200K | $0.2877 | $1.1507 |
| 10 | LongCat-2.0 | Meituan | 1M | $0.3 | $1.2 |
| 11 | qwen3.7-plus | Alibaba | 1M | $0.3077 | $1.2308 |
| 12 | step-1o-turbo-vision | StepFun | 32K | $0.4 | $1.28 |
| 13 | deepseek-v4-pro | DeepSeek | 1M | $0.435 | $0.87 |
| 14 | mimo-v2.5-pro | Xiaomi | 1M | $0.435 | $0.87 |
| 15 | doubao-seed-2-1-turbo-260628 | ByteDance | 256K | $0.4615 | $2.3077 |
| 16 | qwen3.6-plus | Alibaba | 1M | $0.5 | $3 |
| 17 | MiniMax-M2.7-highspeed | MiniMax | 200K | $0.5753 | $2.3014 |
| 18 | doubao-seed-2-1-pro-260628 | ByteDance | 256K | $0.9231 | $4.6154 |
| 19 | kimi-k2.6 | Moonshot | 256K | $0.95 | $3.9462 |
| 20 | kimi-k2.7-code | Moonshot | 256K | $0.95 | $3.9462 |
| 21 | glm-5 | Zhipu AI | 200K | $1 | $3.6667 |
| 22 | glm-5-turbo | Zhipu AI | 200K | $1.2 | $4.4571 |
| 23 | glm-5.2 | Zhipu AI | 1M | $1.4 | $4.4 |
| 24 | glm-5.1 | Zhipu AI | 200K | $1.4 | $4.4 |
| 25 | stepaudio-2.5-chat | StepFun | $1.5 | $3.5 | |
| 26 | kimi-k2.7-code-highspeed | Moonshot | 256K | $1.7808 | $7.3973 |
| 27 | qwen3.8-max | Alibaba | 1M | $1.99 | $5.97 |
| 28 | qwen3.7-max | Alibaba | 1M | $2.5 | $7.5 |
| 29 | kimi-k3 | Moonshot | 1M | $2.7397 | $13.6986 |
| 30 | step-1o-audio | StepFun | $3.85 | $9.24 |
Video generation models
Kling, Seedance, Hailuo, Wan, and HappyHorse.
Video models do not have LLM context windows in the data source. They are billed in USD by the unit shown below. Test the prompt, duration, resolution, reference-control needs, and failure mode before choosing a production route.
| Model | Vendor | Capabilities | Billing |
|---|---|---|---|
| kling-3.0-turbo | Kuaishou | Video, Text-to-Video, Image-to-Video, Audio, Fast | $0.5556 per generation |
| MiniMax-H3 | MiniMax | Video, Text-to-Video, Image-to-Video, Reference-to-Video, Audio, Open Weights | $0.08 per second |
| doubao-seedance-2-5-260628 | ByteDance | Video, Text-to-Video, Image-to-Video | Per generation |
| MiniMax-Hailuo-2.3 | MiniMax | Video, Text-to-Video, Image-to-Video | $0.2778 per generation |
| doubao-seedance-2-0-mini-260615 | ByteDance | Video, Text-to-Video, Fast | Per generation |
| kling-v3 | Kuaishou | Video, Text-to-Video, Image-to-Video, Audio | $0.4799 per 5-second 720p silent generation |
| kling-v3-omni | Kuaishou | Video, Reference-to-Video, Editing | $0.4167 per generation |
| MiniMax-Hailuo-02 | MiniMax | Video, Text-to-Video, Image-to-Video | $0.2778 per generation |
| doubao-seedance-2-0-260128 | ByteDance | Video, Text-to-Video, Image-to-Video | Per generation |
| MiniMax-Hailuo-2.3-Fast | MiniMax | Video, Image-to-Video, Fast | $0.1875 per generation |
| doubao-seedance-2-0-fast-260128 | ByteDance | Video, Text-to-Video, Image-to-Video, Fast | Per generation |
| happyhorse-1.1-t2v | Alibaba | Video, Text-to-Video | $0.075 per second |
| happyhorse-1.1-r2v | Alibaba | Video, Reference-to-Video | $0.075 per second |
| happyhorse-1.1-i2v | Alibaba | Video, Image-to-Video | $0.075 per second |
| wan2.7-t2v | Alibaba | Video, Text-to-Video | $0.1 per second |
| wan2.7-r2v | Alibaba | Video, Reference-to-Video | $0.35 per second |
| wan2.7-videoedit | Alibaba | Video, Video-Edit | $0.2 per second |
| wan2.7-i2v | Alibaba | Video, Image-to-Video, Audio | $0.1 per second |
Image generation models
Seedream and Wan image models.
Image models do not have LLM context windows in the data source. They are billed per image in USD.
| Model | Vendor | Capabilities | Billing |
|---|---|---|---|
| step-2x-large | StepFun | Image, Text-to-Image | $0.0155 per image |
| step-image-edit-2 | StepFun | Image, Image Editing, Fast | $0.0035 per image |
| image-01 | MiniMax | Image, Text-to-Image | $0.0035 per image |
| doubao-seedream-5-0-260128 | ByteDance | Image, Text-to-Image, Image-to-Image, 4K | $0.0306 per image |
| wan2.7-image | Alibaba | Image, Text-to-Image, Image Editing, Multi-Reference | $0.0278 per image |
| wan2.7-image-pro | Alibaba | Image, Text-to-Image, Image Editing, Multi-Reference, 4K | $0.0694 per image |
How should I choose a Chinese AI model?
Start with the route that matches your constraint, then test candidates with your prompts and acceptance criteria. This hub publishes current catalog facts, not a universal quality leaderboard. For independent quality benchmarks, use a third-party source such as Artificial Analysis.
Which models have the longest context?
The 1M-context LLMs in the current data are deepseek-v4-flash, mimo-v2.5, qwen3.6-flash, MiniMax-M3, LongCat-2.0, qwen3.7-plus, deepseek-v4-pro, mimo-v2.5-pro, qwen3.6-plus, glm-5.2, qwen3.8-max, qwen3.7-max, kimi-k3.
Which models are cheapest?
Among token-billed LLMs, step-3.5-flash and step-3.5-flash-2603 share the lowest listed input price at $0.1 per 1M input tokens. The next listed input price is deepseek-v4-flash at $0.14 per 1M input tokens.
Which model has the lowest latency?
ChinaAPI does not publish a cross-region latency ranking. Measure the same prompt, region, concurrency, and streaming mode before using latency as a production decision.
Start testing
Use the decision routes as a shortlist, then test your own workload.
Catalog facts narrow the field. Final model choice should still come from your prompts, data, latency needs, acceptance criteria, and total cost per accepted result.
Try ChinaAPI
Start with free credit or review live pricing before routing production traffic.