Chinese model releases to evaluate when teams need model weights, local deployment, inference control or lower vendor lock-in.
Quick answers
Decision standard
Prioritize products with public model repositories, documented licenses, deployment notes or explicit self-hosting paths.
DeepSeek
The English docs now make DeepSeek's current API shape explicit: V4-Pro/V4-Flash, dual OpenAI/Anthropic compatibility, thinking mode, tool calls, caching and agent-tool integrations.
Best fit · Developers evaluating Chinese model APIs for coding agents, reasoning, tool use and OpenAI/Anthropic-compatible migration.
Coverage · 100/100
OpenBMB
OpenBMB belongs in LLM & API as a Chinese open-source foundation-model ecosystem, with MiniCPM and MiniCPM-V as the main product signals tracked here.
Best fit · Researchers and engineering teams evaluating Chinese open-source models for edge inference, multimodal understanding, RAG and agent experiments.
Coverage · 100/100
AtomGit
AtomCode belongs in AI Coding because it is a full terminal coding agent and Claude Code alternative, not a generic chat assistant or raw model API.
Best fit · Developers who want an open-source terminal coding agent for Chinese and global model APIs, especially DeepSeek, GLM, Qwen and local Ollama workflows.
Coverage · 100/100
Alibaba Cloud
Qwen Cloud makes Qwen available for global evaluation through an English marketplace, docs, pricing and compatible API paths.
Best fit · Developers evaluating Qwen3.7, Qwen Cloud APIs, coding agents and multimodal Chinese model coverage from an English international platform.
Coverage · 100/100
Alibaba Cloud
Qwen image models are visible in the international marketplace and deserve a separate image-category entry.
Best fit · Teams evaluating Chinese image-generation APIs with English docs and Qwen Cloud billing.
Coverage · 100/100
Alibaba Cloud
Qwen Cloud has enough official audio evidence to warrant a separate audio-category profile.
Best fit · Teams evaluating Chinese speech synthesis, voice cloning, ASR and realtime speech APIs through an English platform.
Coverage · 100/100
MiniMax
M3 should be tracked separately because the official homepage now gives it a distinct model page and positions it as the current frontier coding model.
Best fit · Developers evaluating a China-origin frontier model for coding, long-context agentic work and multimodal reasoning.
Coverage · 100/100
Alibaba Cloud
Qwen Code is a distinct AI coding product and should be tracked separately from Qwen Cloud models or token plans.
Best fit · Developers who want a terminal coding agent for Qwen models and prefer open-source tooling with editor integrations.
Coverage · 100/100
Alibaba Cloud
Qwen-AgentWorld is tracked separately from Qwen Code because its public surface is a language world model plus AgentWorldBench, not a terminal coding product.
Best fit · Researchers and agent builders who need a simulator or benchmark for tool, terminal, SWE, Android, web and OS agent environments.
Coverage · 100/100
DeepReinforce
Ornith 1.0 belongs in AI Coding because its public model cards and collection position it around agentic coding benchmarks and coding-agent deployment.
Best fit · Teams evaluating open agentic-coding models for self-hosted coding agents, terminal automation, SWE-Bench workflows and long-context tool use.
Coverage · 100/100
JD JoyAI
JoyAI-VL-Interaction deserves a separate profile because it releases a full real-time interaction model and system stack, not just a static VLM checkpoint.
Best fit · Researchers and product teams exploring proactive webcam, livestream, monitoring, cooking guidance, live commentary and event-driven visual interaction systems.
Coverage · 100/100
Moonshot AI
Kimi K3 merits a separate LLM & API profile because it is a flagship model with its own architecture, context length, hosted pricing and public open-weight release, rather than a minor K2.7 Code revision.
Best fit · Teams evaluating a frontier-scale Chinese model for repository-scale coding, multimodal technical work, agentic research and long-context knowledge workflows.
Coverage · 100/100
Moonshot AI
K2.7 Code has its own open weights, architecture, benchmarks, API pricing and Kimi Code default-model behavior, so it deserves a dedicated model profile separate from Kimi Code and the broader Kimi API.
Best fit · Teams evaluating open-weight coding models for repository-scale agents, Kimi Code workflows or OpenAI-compatible API integration.
Coverage · 100/100
Moonshot AI
The public repository confirms that Kimi Code is an installable MIT-licensed agent platform, while the K2.7 Code resource confirms its current default hosted coding model.
Best fit · Developers who want a fast open terminal agent with multimodal coding context, parallel subagents and editor integration.
Coverage · 100/100
Zhipu AI
Z.ai now has an English product surface, while BigModel remains the API evidence base for the full GLM product line.
Best fit · Developers comparing Chinese multimodal model APIs, agent services and OpenAI-compatible migration paths.
Coverage · 100/100
Ant Group
Ring has a coding-agent and tool-use profile that fits AI Coding better than the generic Ant Ling entry.
Best fit · Teams comparing Chinese reasoning and coding-agent models for tool orchestration, repository work and long-horizon autonomous execution.
Coverage · 100/100
MiniMax
The international docs show MiniMax as a full multimodal API platform rather than only a Hailuo video product.
Best fit · Developers who want one China-origin platform for coding models, speech, video, image, music and multimodal agent tooling.
Coverage · 100/100
MiniMax
Mavis is a productized multi-agent runtime with background execution, verification and recoverable state, rather than only a general chat assistant.
Best fit · Teams delegating long-running coding, research and document workflows that benefit from parallel roles and independent verification.
Coverage · 100/100
Meituan LongCat
LongCat-AudioDiT belongs in AI Audio because it is a direct-text-to-speech and voice-cloning model with released code and weights, not a generic research paper.
Best fit · Researchers and speech teams evaluating open-source TTS, waveform-latent diffusion and zero-shot voice cloning.
Coverage · 100/100 · backfill: freshness
NetEase Youdao
Confucius4-TTS belongs in AI Audio because it is an open TTS and voice-cloning engine with model downloads, inference code and benchmark claims.
Best fit · Speech researchers and localization teams evaluating open multilingual TTS, cross-lingual dubbing, zero-shot voice cloning and emotion-preserving voice transfer.
Coverage · 100/100
Baidu AI Cloud
Baidu now has English model communications through the ERNIE Blog, while Qianfan remains the main platform when enterprise platform, agent orchestration and China-cloud deployment matter.
Best fit · Enterprises and developers already evaluating Baidu Cloud, China-local deployment, agent platforms or ERNIE multimodal models.
Coverage · 100/100
Tencent Cloud
Hunyuan is strategically important because Tencent combines cloud APIs, global cloud billing, a broad open-source model organization and a newly global 3D creation engine.
Best fit · Developers and enterprises comparing Chinese model APIs, Tencent Cloud integration, 3D asset generation and open-source model deployment.
Coverage · 100/100
Tencent Cloud
Hy3-preview is the English-documented Hunyuan open LLM profile.
Best fit · Teams evaluating open Chinese reasoning and agent models with long-context deployment requirements.
Coverage · 100/100
Tencent Cloud
Hunyuan 3D Global is Tencent Hunyuan's international consumer/creator product entry.
Best fit · Creators and enterprises needing Chinese AI 3D asset generation with a global web product and Tencent Cloud API path.
Coverage · 100/100
Tencent Cloud
The open-source organization helps show Hunyuan's model breadth beyond Tencent Cloud APIs.
Best fit · Developers surveying Tencent's open-source model assets across LLM, video, image, 3D and OCR.
Coverage · 100/100
AgentScope AI
AgentTeams belongs in Productivity because its core value is collaborative agent orchestration, not model hosting or a narrow coding workflow.
Best fit · Teams that need a self-hosted, auditable multi-agent operating layer for enterprise collaboration and task orchestration.
Coverage · 100/100
AgentScope AI
QwenPaw belongs in Productivity because it is a worker runtime for collaborative agent orchestration, not a direct model API or a full coding IDE.
Best fit · Teams that want a browser-friendly worker runtime inside a transparent multi-agent room rather than a monolithic assistant.
Coverage · 100/100 · backfill: sources
ByteDance / Volcano Engine
DeerFlow belongs in Productivity because it is a SuperAgent harness broader than an AI coding agent: its core promise is long-horizon research, creation and workflow execution through skills, sub-agents and sandboxes.
Best fit · Teams that want to self-host a Chinese-origin open-source agent harness for deep research, report generation, coding, file-based workflows and long-running multi-agent tasks.
Coverage · 100/100
iFlytek
Astron expands iFlytek coverage from Spark model access into agentic application orchestration and open-source workflow tooling.
Best fit · Teams evaluating an iFlytek-linked open-source and hosted platform for agent workflows, RAG applications and enterprise automation.
Coverage · 100/100
StepFun
StepFun is important because it combines multimodal model depth, open-source releases and device commercialization, but overseas usability still needs hands-on checks.
Best fit · Teams evaluating Chinese multimodal models, open-source agent models, video/audio generation or device-side AI partnerships.
Coverage · 100/100
Xiaomi MiMo
MiMo matters because Xiaomi is exposing an English-facing model platform across web, API, AI Studio and open-source channels.
Best fit · Developers comparing Chinese agent models, multimodal model families, speech models, English web demos and open-source deployment options.
Coverage · 100/100
Xiaomi MiMo
MiMo-V2-Flash gives MiMo a concrete English-documented open-weight anchor with price, context, test and deployment details.
Best fit · Teams evaluating a low-cost open-weight Chinese model for coding agents, long-context tasks and API/self-hosted comparison.
Coverage · 100/100
Xiaomi MiMo
MiMo now has enough English-facing speech signals to deserve a separate audio profile.
Best fit · Teams watching Xiaomi's speech stack for ASR, TTS and voice-agent experiments.
Coverage · 100/100 · backfill: pricing
Xiaomi MiMo
Separating the platform from model cards makes MiMo easier to compare with Qwen Cloud, DeepSeek Platform and Kimi Platform.
Best fit · Developers who want to test MiMo through hosted API and AI Studio before deciding on self-hosting.
Coverage · 100/100
Xiaomi MiMo
MiMoCode is a distinct coding-agent product rather than another MiMo model card: it packages memory, orchestration, tools and provider routing into an installable open-source CLI.
Best fit · Developers who want an open terminal coding agent focused on durable project memory, long-running tasks and configurable model providers.
Coverage · 100/100
DeepSeek: DeepSeek is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
OpenBMB / MiniCPM: OpenBMB / MiniCPM is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
AtomCode: AtomCode is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
Qwen: Qwen is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
Qwen-Image / Wan Image: Qwen-Image / Wan Image is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
Qwen Audio / CosyVoice: Qwen Audio / CosyVoice is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
MiniMax M3: MiniMax M3 is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
Qwen Code: Qwen Code is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
Qwen-AgentWorld: Qwen-AgentWorld is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
Ornith 1.0: Ornith 1.0 is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
JoyAI-VL-Interaction: JoyAI-VL-Interaction is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
Kimi K3: Kimi K3 is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
Kimi K2.7 Code: Kimi K2.7 Code is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
Kimi Code: Kimi Code is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
Z.ai BigModel / GLM: Z.ai BigModel / GLM is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
Ring: Ring is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
MiniMax API Platform: MiniMax API Platform is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
MiniMax Agent / Mavis: MiniMax Agent / Mavis is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
LongCat-AudioDiT: LongCat-AudioDiT is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
Confucius4-TTS: Confucius4-TTS is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
ERNIE / Baidu Qianfan: ERNIE / Baidu Qianfan is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
Tencent Hunyuan: Tencent Hunyuan is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
Hy3 Preview: Hy3 Preview is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
Hunyuan 3D Global: Hunyuan 3D Global is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
Tencent Hunyuan Open Models: Tencent Hunyuan Open Models is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
AgentTeams (formerly HiClaw): AgentTeams (formerly HiClaw) is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
QwenPaw: QwenPaw is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
DeerFlow: DeerFlow is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
Astron Agent: Astron Agent is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
StepFun / Step: StepFun / Step is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
Xiaomi MiMo: Xiaomi MiMo is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
MiMo-V2-Flash: MiMo-V2-Flash is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
MiMo Speech Models: MiMo Speech Models is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
MiMo API Platform: MiMo API Platform is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.
MiMoCode: MiMoCode is included because its sources mention open models, model hubs, self-hosting or deployment-oriented evidence.