Qwen3.5 122B A10B (Alibaba Cloud)

The LLM-Stats MVBench leaderboard ranks Qwen3.5-122B-A10B second with a score of 0.766 across 18 evaluated multimodal models, behind GLM-5.3-Flash at 0.778. The benchmark covers 20 video tasks requiring temporal reasoning, perception, and spatial understanding beyond single-frame analysis. Qwen3.5-122B-A10B is listed at 122B parameters with a 262K context window and reported pricing of $0.29 input / $2.40 output per 1M tokens. The score puts it ahead of Qwen3.6-27B (0.755) and Qwen3.5-35B-A3B (0.748) on multimodal video comprehension.

The LLM-Stats MMBench-V1.1 bilingual multimodal leaderboard places Qwen3.5-122B-A10B tied for rank 1 with a score of 0.928 across 20 evaluated vision-language models, level with sibling Qwen3.6-35B-A3B. MMBench-V1.1 evaluates multimodal capabilities via multiple-choice questions in English and Chinese. Qwen3.5-122B-A10B is reported at 122B parameters with a 262K context window and listed pricing of $0.29 input / $2.40 output per 1M tokens. It edges out Qwen3.5-27B (0.926) and Qwen3.6-27B (0.923) on the vision-language evaluation.

The LLM-Stats MMStar leaderboard ranks Qwen3.5-122B-A10B second with a score of 0.829 across 27 evaluated multimodal models, trailing only Qwen3.6 Plus at 0.833. MMStar is a vision-language benchmark assessing multimodal capabilities via curated multiple-choice evaluation items. Qwen3.5-122B-A10B is listed at 122B parameters with a 262K context window and reported pricing of $0.29 input / $2.40 output per 1M tokens. It places ahead of Qwen3.5-35B-A3B (0.819), Qwen3.6-27B (0.814), and ByteDance's Seed 1.8 (0.799) on the multimodal benchmark.

On the MLVU multi-task long video understanding leaderboard, Qwen3.5-122B-A10B places second with a score of 0.873, narrowly behind Qwen3.7-Plus at 0.874. The benchmark evaluates multimodal models on videos ranging from 3 minutes to 2 hours across nine tasks including reasoning, captioning, and summarization. The MLVU result, covering 10 evaluated models, shows Qwen3.5-122B-A10B (122B, 262K context) outperforming Qwen3.6-27B (0.866) and Qwen3.6-35B-A3B (0.862) on long-video comprehension. This indicates strong multimodal long-context performance for the 122B-A10B variant within the Qwen family.

On the BrowseComp-zh leaderboard, Qwen3.5-122B-A10B ranks third with a score of 0.699, behind ByteDance's Seed 1.8 (0.813) and Qwen3.5-397B-A17B (0.703). The benchmark comprises 289 multi-hop Chinese-web agent questions spanning 11 domains including Technology, Film & TV, and Medicine. Qwen3.5-122B-A10B is listed at 122B parameters with a 262K context window and reported pricing of $0.29 input / $2.40 output per 1M tokens on BrowseComp-zh. The score places it above larger peers like DeepSeek-V3.2 Thinking (0.650) and GLM-4.7 (0.666) on Chinese-web reasoning tasks.

Alibaba's Qwen team released the Qwen3.5 Medium Model series in late February 2026, with the open-source Qwen3.5-122B-A10B among three variants available for commercial use under Apache 2.0 on Hugging Face and ModelScope. The release also includes a fourth proprietary Qwen3.5-Flash accessible only via the Alibaba Cloud Model Studio API. Qwen3.5-122B-A10B is built on a hybrid architecture combining Gated Delta Networks with a sparse Mixture-of-Experts system, and is engineered to stay accurate under 4-bit weight and KV cache quantization. The models target performance comparable to Claude Sonnet 4.5 and GPT-5-mini on third-party benchmarks while remaining runnable on consumer-grade GPUs.

The LLMBoard PMC-VQA medical visual question answering leaderboard places Qwen3.5-122B-A10B at rank 1 with a score of 63.30%, earning the 100th percentile among three evaluated models as of September 27, 2026. PMC-VQA is a biomedical VQA benchmark built on medical literature and figures. Qwen3.5-122B-A10B from Alibaba Cloud / Qwen Team leads its two Qwen3.5 siblings, Qwen3.5-27B (62.40%) and Qwen3.5-35B-A3B (62.00%), on medical multimodal question answering. The leaderboard lists reported pricing of $0.40 input / $3.20 output per 1M tokens for the model.

The LLMBoard MVBench leaderboard places Qwen3.5-122B-A10B from Alibaba Cloud / Qwen Team at rank 2 with a score of 76.60%, putting it in the 94.12 percentile across 18 evaluated models as of September 27, 2026. It trails only GLM-5.3-Flash from Zhipu AI, which leads at 77.80%. The multimodal video MVBench benchmark covers 20 tasks requiring temporal understanding beyond single-frame analysis, and ranks Qwen3.5-122B-A10B ahead of siblings like Qwen3.5-35B-A3B (74.80%) and Qwen3.5-27B (74.60%). The result confirms strong video-understanding performance for the 122B-A10B variant within the Qwen3.5 family.