Unverified50% confidenceBenchmarkExact time
The Qwen2.5-VL series performs especially well on the DocVQA and OCRBench benchmarks.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/31/2026
First Seen
Valid until: 11/29/2026
Sources
Related Entities
Related Claims
UnverifiedQwen2.5系列在后训练阶段引入了大量工具调用对齐数据，并通过RLHF和DPO强化了结构化输出的可靠性70% similarUnverifiedThe Qwen2.5-VL series released by Alibaba's Tongyi Qianwen team includes a 7B version that performs excellently on tasks such as image understanding, document parsing, and OCR.69% similarUnverifiedQwythos-9B is built on top of the uncensored Qwen 3.5 base model.68% similarUnverified阿里巴巴的Qwen2.5在多项基准测试中表现优异，Mistral以其高效的Mixture of Experts架构著称66% similarUnverifiedSeries such as Qwen2.5 and Gemma3 have already released 1B–3B quantized versions optimized specifically for edge scenarios.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/833037API
curl https://kongchang.com/api/v1/knowledge/claims/833037MCP
get_claim(id=833037)