Model dossiers · 26
The Chinese multimodal model library
Every model gets a dossier: eight dimension scores, public benchmarks, scenario mapping, pricing, licence and alternatives. Browse by tier or search the name.
Qwen3.8-Max
Alibaba · 1M (256K native)
The flagship that changed who is #1
Doubao Seed 2.0 Pro
ByteDance · 128K
The practical ceiling for Chinese multimodal work
Qwen3.5-397B-A17B
Alibaba · 256K (up to 1M)
The multimodal flagship that behaves like infrastructure
Qwen3-Omni-30B
Alibaba · 32K
The open option that genuinely unifies audio
ERNIE 5.1
Baidu · 128K
Freshness through search grounding
DeepSeek-V4.1-Flash
DeepSeek · 128K
The nail that sells document skill for ¥0.6
Qwen3.8-Flash
Alibaba · 256K
An A-tier ticket for ¥1.2
Keye-VL-2.0
Kuaishou · 256K
The open specialist for long-video timing
Hunyuan-Vision
Tencent · 128K
The safe choice inside the Tencent stack
InternVL 4.0
Shanghai AI Laboratory · 256K
The open-source perception ceiling, with video as the priced-in trade
DeepSeek-OCR 2
DeepSeek · 64K
A specialist that makes OCR nearly lossless
Pangu 5.5
Huawei · 128K
An industry brain for manufacturing, energy and transport
Ant Group Ling 2
Ant Group · 256K
The finance-scenario model with Chinese 92
WITA-Omni Preview
AgiBot · Undisclosed
An omni brain that reads the room
Emu3.5
BAAI · 128K
A world model built on next-state prediction
Zidong Taichu 2.0
CASIA · 128K
The academic all-modality entry: more papers than product
MiniCPM-V 4.5
ModelBest · 96K
A phone-sized Chinese OCR workhorse
Hikvision Guanlan 3
Hikvision · 128K
The only multimodal model trained for surveillance video
JD Yanxi 2
JD.com · 128K
The retail specialist for product images and supply-chain documents
NetEase Fuxi 2
NetEase · 128K
The multimodal player for games and content
NeuroXess-N1
NeuroXess · 128K
The only Chinese multimodal model that takes neural signals as an input modality
BAGEL-2
ByteDance · PKU · 64K
An open unified model balancing understanding and generation
Cloudwalk Congrong 2
Cloudwalk · 128K
The only CV “little dragon” still shipping multimodal
Vendor map
Who is building native multimodal models
Internet giants cover general capability; startups differentiate on long context, omni-modality and edge; research labs push architecture.
About the model library
What does the evidence grade mean?
A means third-party reproduction or cross-verification across authoritative boards; B is mainly official reports with partial community reproduction; C is a single source. The grade does not change the score, but it changes how much you should trust it.
Why several models from one vendor?
Flagships, open variants and specialists serve different jobs. Qwen3.5 is a general flagship, Qwen3-Omni targets omni prototypes, and DeepSeek-OCR 2 only does documents. Listing them together is closer to real procurement than flagships alone.
Why do prices differ from vendor pages?
We use public list prices or convertible self-host estimates, excluding tier discounts, cache hits, batch deals and credits. Always confirm current vendor pricing before purchase.
Personal ranking · Not an official score
Every ranking, score and comparison on this page is an independent personal ranking, not an official score, and represents no vendor, institution or regulator. Scores are a subjective mapping of public evidence under our own weights — change the weights and the order changes. Disagree, or spotted a factual error? Contact the site owner: jizhi2024@foxmail.com — we check each report and log corrections in the changelog.