Supported Models

Get the latest info here: https://github.com/vllm-project/vllm-ascend/issues/1608

Note

This list is based on the vLLM main community model list and categorizes the support status of vllm-ascend with labels such as Supported, Experimental, and Unmaintained. Models not listed may be verified independently.

Legend Description:

Legend	Description	Usage	Recommendation
✅	Supported	Fully implemented and verified, meeting both functional and performance standards.	Can be used in production environments.
🔵	Experimental	Implemented and basically functional, but still under improvement. May have instability or performance fluctuations.	Can be used for testing/pre-research, production use requires risk assessment.
🟡	Unverified	Code has been merged but not yet subjected to system testing. Behavior is uncertain and cannot be evaluated.	Users assume the risk of use.
🔴	Unmaintained	Previously supported, but now discontinued and no longer maintained. Issues will not be fixed.	Do not use in new projects.
❌	Unsupported	Not implemented in the current version and cannot be used.	Not available.

Text-Only Language Models

Generative Models

Core Supported Models

950PR&950DT ProductsA2/A3Atlas 300I DUO

Model	Support	Note	BF16	Supported Hardware	W8A8	Chunked Prefill	Automatic Prefix Cache	LoRA	Speculative Decoding	Async Scheduling	Tensor Parallel	Pipeline Parallel	Expert Parallelism	Data Parallel	Prefill-decode Disaggregation	Piecewise AclGraph	Fullgraph AclGraph	max-model-len	MLP Weight Prefetch	Doc
DeepSeek V4-Flash	✅	Native mixed MXFP8/MXFP4 weights		950DT Products	✅	✅	✅		✅	✅		✅	✅	✅	✅		✅	1M		DeepSeek V4-Flash
DeepSeek V4-Pro	✅	Native mixed MXFP8/MXFP4 weights		950DT Products	✅	✅	✅		✅	✅		✅	✅	✅	✅		✅	1M		DeepSeek V4-Pro
DeepSeek-V3.1	🔴		✅	950DT Products	✅	✅	✅		✅	✅	✅		✅	✅	✅	✅	✅	240k		DeepSeek-V3.1
GLM-5.1	✅		✅	950DT Products	✅	✅	✅		✅	✅	✅		✅	✅	✅		✅	200k		GLM-5.1
GLM-5.2	✅	W8A8C8 MXFP8 / W4A4C8 MXFP4	🟡	950DT Products	✅	🟡	✅		✅	✅	✅	✅	✅	✅	✅	🟡	✅	🟡		GLM-5.2
GLM-5.3	✅	W8A8C8 MXFP8 / W4A4C8 MXFP4	🟡	950DT Products	✅	🟡	✅		✅	✅	✅	✅	✅	✅	✅	🟡	✅	1M		GLM-5.3

Model	Support	Note	BF16	Supported Hardware	W8A8	Chunked Prefill	Automatic Prefix Cache	LoRA	Speculative Decoding	Async Scheduling	Tensor Parallel	Pipeline Parallel	Expert Parallelism	Data Parallel	Prefill-decode Disaggregation	Piecewise AclGraph	Fullgraph AclGraph	max-model-len	Doc
DeepSeek V4-Flash	✅		✅	A2/A3	✅	✅	✅		✅	✅	✅		✅	✅	✅		✅	1M	DeepSeek-V4-Flash
DeepSeek V4-Pro	✅		✅	A2/A3	✅	✅	✅		✅	✅	✅		✅	✅	✅		✅	1M	DeepSeek-V4-Pro
DeepSeek V3/3.1	🔴		✅	A2/A3	✅	✅	✅		✅		✅	✅	✅	✅	✅	✅	✅	240k	DeepSeek-V3.1
DeepSeek V3.2	🔴		✅	A2/A3	✅	✅	✅	✅	✅	✅	✅	✅	✅	✅	✅	✅	✅	160k	DeepSeek-V3.2
DeepSeek R1	🔴		✅	A2/A3	✅	✅	✅		✅		✅	✅	✅	✅	✅	✅	✅	128k	DeepSeek-R1
Qwen3-Dense	✅		✅	A2/A3	✅	✅	✅			✅	✅			✅		✅	✅	128k	Qwen3-Dense
Qwen3-30B-A3B	✅		✅	A2/A3	✅	✅	✅		✅	✅	✅		✅	✅		✅	✅		Qwen3-30B-A3B
Qwen3-Coder-30B-A3B	🔴		✅	A2/A3	✅	✅	✅		✅	✅	✅		✅	✅		✅	✅		Qwen3-Coder-30B-A3B
Qwen3-235B-A22B	🔴		✅	A2/A3	✅	✅	✅			✅	✅		✅	✅	✅	✅	✅	256k	Qwen3-235B-A22B
Qwen3-Next	🔴		✅	A2/A3	✅						✅			✅		✅	✅		Qwen3-Next
GLM-4.x	🔴			A2/A3	✅	✅	✅		✅	✅	✅		✅	✅	✅	✅	✅	198k	GLM-4.x
GLM-5	🔴		✅	A2/A3	✅	✅	✅		✅	✅	✅	✅	✅	✅	✅	✅	✅	200k	GLM-5
GLM-5.1	✅		✅	A2/A3	✅	✅	✅		✅	✅	✅	✅	✅	✅	✅	✅	✅	200k	GLM-5.1
GLM-5.2	✅	W8A8C8	✅	A2/A3	✅	✅	✅		✅	✅	✅	✅	✅	✅	✅	✅	✅	200k	GLM-5.2
GLM-5.3	✅	W8A8C8	✅	A2/A3	✅	✅	✅		✅	✅	✅	✅	✅	✅	✅	✅	✅	135k (A2) / 1M (A3)	GLM-5.3
Gemma4	🔵		✅	A2/A3/950PR&950DT Products		✅	✅			✅	✅			✅		✅	✅		Gemma4
Kimi-K2-Thinking	🔴			A2/A3															Kimi-K2-Thinking
DeepSeekOCR2	🔴		✅	A2/A3		✅				✅									DeepSeekOCR2
MiniMax-M2.5/2.7	🔴		✅	A2/A3/Ascend950 (Ascend950 experimental)	✅	✅	✅	❌	✅	✅	✅	🟡	✅	✅	✅	🟡	✅	200k	MiniMax-M2
Qwen2.5-Math-RM-72B	🔴	vllm-rm, tensor_parallel_size=4, max_model_len=4096	✅	A2	✅	🟡	🟡	❌	🟡	✅	✅	🟡	🟡	🟡	🟡	🟡	🟡	4096	Qwen2.5-Math-RM-72B
Hy3-preview	🔴			A3															Hy3-preview
Hy4-preview (Experimental)	🔵			A3															Hy4-preview (Experimental)

Model	Support	Note	BF16	Supported Hardware	W8A8	Chunked Prefill	Automatic Prefix Cache	LoRA	Speculative Decoding	Async Scheduling	Tensor Parallel	Prefill-decode Disaggregation	Piecewise AclGraph	Fullgraph AclGraph	max-model-len	Doc
Qwen3-Dense	✅		❌	Atlas 300I DUO	✅	✅	✅	❌	🟡	✅	✅	❌	✅	✅	20k	Qwen3-Dense
Qwen3-30B-A3B	✅		❌	Atlas 300I DUO	✅	✅	✅	❌	🟡	✅	✅	❌	✅	✅	16k	Qwen3-30B-A3B

Pooling Models

950PR&950DT ProductsA2/A3Atlas 300I DUO

Model	Support	Supported Hardware	Doc
Qwen3-Embedding	🔴	Ascend 950DT&950PR Products	Qwen3-Embedding
Qwen3-VL-Embedding	🔴	Ascend 950DT&950PR Products	Qwen3-VL-Embedding
Qwen3-Reranker	🔴	Ascend 950DT&950PR Products	Qwen3-Reranker
Qwen3-VL-Reranker	🔴	Ascend 950DT&950PR Products	Qwen3-VL-Reranker

Model	Support	Note	Supported Hardware	W8A8	Doc
Qwen3-Embedding	🔴		A2/A3	🟡	Qwen3-Embedding
Qwen3-VL-Embedding	🔴		A2/A3	🔵	Qwen3-VL-Embedding
Qwen3-Reranker	🔴		A2/A3	🟡	Qwen3-Reranker
Qwen3-VL-Reranker	🔴		A2/A3	🔵	Qwen3-VL-Reranker
Qwen2.5-Math-RM-72B	🔴	Reward Model, gsm8k_correctness accuracy=0.80	A2	Qwen2.5-Math-RM-72B

Model	Support	Note	Supported Hardware	W8A8	Doc
Qwen3-Embedding	🔴	FP16	Atlas 300I DUO	🟡	Qwen3_Embedding
Qwen3-VL-Embedding	🔴	FP16	Atlas 300I DUO	🔵	Qwen3_VL_Embedding
SigLIP2	✅	FP16; text/image embedding	Atlas 300I DUO	🔵	SigLIP2
Qwen3-Reranker	🔴	FP16	Atlas 300I DUO	🟡	Qwen3_Reranker
Qwen3-VL-Reranker	🔴	FP16	Atlas 300I DUO	🔵	Qwen3_VL_Reranker

Multimodal Language Models

Generative Models

Core Supported Models

950PR&950DT ProductsA2/A3Atlas 300I DUO

Model	Support	Note	BF16	Supported Hardware	W8A8	Chunked Prefill	Automatic Prefix Cache	LoRA	Speculative Decoding	Async Scheduling	Tensor Parallel	Pipeline Parallel	Expert Parallelism	Data Parallel	Prefill-decode Disaggregation	Piecewise AclGraph	Fullgraph AclGraph	max-model-len	MLP Weight Prefetch	Doc
GLM-5.3-Flash	✅	Text, image, video	✅	950DT Products	✅	✅	✅		✅	✅	✅	✅	✅	✅	✅	✅	✅	133120		GLM-5.3-Flash
Qwen3.5-397B-A17B	✅	Text, image, video	✅	950DT Products	✅	✅	✅		✅	✅	✅		✅	✅	✅	✅	✅	1010000		Qwen3.5-397B-A17B
Qwen3.6-27B	✅		✅	950DT Products	✅	✅	✅		✅	✅	✅		✅	✅	✅	✅	✅	262144		Qwen3.5-27B / Qwen3.6-27B
Qwen3.6-35B-A3B	✅		✅	950DT Products	✅	✅	✅		✅	✅	✅		✅	✅	❌	✅	✅	135168		Qwen3.6-35B-A3B
Kimi-K3	✅	W4A8; DSpark; FULL_DECODE_ONLY; KV Cache Pool		950DT Products			✅		✅		✅		✅	✅	✅		✅	131072		Kimi-K3
MiniMax-M3	✅	MXFP8; text/image/video		950DT Products		✅	✅		✅	✅	✅	✅	✅	✅	✅	🟡	✅	1M		MiniMax-M3
Step-5-Preview	🔵	Experimental		950DT Products	✅	✅	✅		✅	✅	✅	✅	✅	✅			✅	1M		Step-5-Preview
MiniMax-M3.1-Flash	🔵	Experimental		950DT products	✅	✅	✅		✅	✅	✅	✅	✅	✅	✅	🟡	✅	1M		MiniMax-M3.1-Flash

Model	Support	Note	BF16	Supported Hardware	W8A8	Chunked Prefill	Automatic Prefix Cache	LoRA	Speculative Decoding	Async Scheduling	Tensor Parallel	Pipeline Parallel	Expert Parallelism	Data Parallel	Prefill-decode Disaggregation	Piecewise AclGraph	Fullgraph AclGraph	max-model-len	Doc
GLM-5.3-Flash	✅	Text, image, video	✅	A2/A3	✅	✅	✅		✅	✅	✅	✅	✅	✅	✅	✅	✅	200k (A2) / 1M (A3)	GLM-5.3-Flash
Qwen3-VL	✅			A2/A3							✅					✅	✅		Qwen-VL-Dense
Qwen3-VL-30B-A3B	✅		✅	A2/A3	✅	✅	✅			✅	✅		✅	✅	✅	✅	✅	262144	Qwen3-VL-30B-A3B
Qwen3-VL-235B-A22B	🔴		✅	A2/A3	✅	✅	✅			✅	✅		✅	✅	✅	✅	✅	262144	Qwen3-VL-235B-A22B
DeepSeek-V4.1-Flash	✅	W8A8; 2-node A3 or 4-node A2 colocated deployment	🟡	A2/A3	✅	🟡	✅	❌	✅	🟡	✅	❌	✅	✅	❌	❌	✅	1048576	DeepSeek-V4.1-Flash
Qwen3.5-397B-A17B	✅	Text, image, video	✅	A2/A3	✅	✅	✅		✅	✅	✅		✅	✅	✅	✅	✅	1010000	Qwen3.5-397B-A17B
Qwen3.5-27B / Qwen3.6-27B	✅		✅	A2/A3	✅	✅	✅		✅	✅	✅		✅	✅	✅	✅	✅	262144	Qwen3.5-27B / Qwen3.6-27B
Qwen3.6-35B-A3B	✅		✅	A2/A3	✅	✅	✅		✅	✅	✅		✅	✅	❌	✅	✅	135168	Qwen3.6-35B-A3B
Qwen3.8-Flash-Next	🔵		✅	A3	✅	✅	✅		✅	✅	✅		✅	✅	✅	✅	✅	262144	Qwen3.8-Flash-Next
Qwen3-Omni-30B-A3B-Thinking	🔴			A2/A3							✅		✅						Qwen3-Omni-30B-A3B-Thinking
Kimi-K2.5	🔴			A2/A3		✅	✅		✅	✅	✅		✅	✅	✅	✅	✅	262144	Kimi-K2.5
Kimi-K2.6	✅			A2/A3		✅	✅		✅	✅	✅		✅	✅	✅	✅	✅	262144	Kimi-K2.6
Kimi-K3	✅	W4A8; DSpark; FULL_DECODE_ONLY; KV Cache Pool		A3			✅		✅		✅		✅	✅	✅		✅	131072	Kimi-K3
MiniMax-M3	✅	Text, image, video, and ViT DP	✅	A2/A3	✅	✅	✅		✅	✅	✅	✅	✅	✅	✅	🟡	✅	1M	MiniMax-M3
MiniMax-M3.1-Flash	🔵	Experimental		A2/A3	✅	✅	✅		✅	✅	✅	✅	✅	✅	✅	🟡	✅	1M	MiniMax-M3.1-Flash
Cohere Transcribe	🔵	2B Conformer encoder-decoder ASR, 14 languages (03-2026 & arabic-07-2026)	✅	A2/A3	❌			❌	❌		🟡				❌	🟡	🟡		Cohere-Transcribe
PaddleOCR-VL	🔴			A2/A3															PaddleOCR-VL

Model	Support	Note	BF16	Supported Hardware	W8A8	Chunked Prefill	Automatic Prefix Cache	LoRA	Speculative Decoding	Async Scheduling	Tensor Parallel	Prefill-decode Disaggregation	Piecewise AclGraph	Fullgraph AclGraph	max-model-len	Doc
Qwen3-VL	✅		❌	Atlas 300I DUO	✅	✅	✅	❌	🟡	✅	✅	❌	✅	✅	16k	Qwen-VL-Dense
Qwen3.5-Dense	✅		❌	Atlas 300I DUO	✅	✅	✅	❌	✅	✅	✅	❌	✅	✅	256k	Qwen3.5-Dense
Qwen3.5-35B-A3B	✅		❌	Atlas 300I DUO	✅	✅	✅	❌	✅	✅	✅	❌	✅	✅	256k	Qwen3.5-35B-A3B
Qwen3.5-27B	✅		❌	Atlas 300I DUO	✅	✅	✅	❌	✅	✅	✅	❌	✅	✅	256k	Qwen3.5-27B
Qwen3.6-27B	✅		❌	Atlas 300I DUO	✅	✅	✅	❌	✅	✅	✅	❌	✅	✅	256k	Qwen3.6-27B
Qwen3.6-35B-A3B	✅		❌	Atlas 300I DUO	✅	✅	✅	❌	✅	✅	✅	❌	✅	✅	256k	Qwen3.6-35B-A3B
Qwen3.8-27B	✅		❌	Atlas 300I DUO	✅	✅	✅	❌	✅	✅	✅	❌	✅	✅	256k	Qwen3.8-27B
PaddleOCR-VL	🔴		❌	Atlas 300I DUO	❌	✅	✅	❌	❌	✅	❌	❌	✅	✅	16k	PaddleOCR-VL
Qwen3-ASR	🔴		❌	Atlas 300I DUO	❌	✅	✅	❌	❌	✅	🟡	❌	✅	✅	4096	Qwen3-ASR-1.7B