🤖
Qwen 3.5 MoE 35B 指令模式詢問
Reddit 用戶因辦公室頻寬限制，尋求 Qwen 3.5 MoE 35B 不含推理鏈的指令模式測試結果。對 Qwen 在 2507 發布後重返混合推理模型感到驚訝。社群討論持續中。
Search
Tag: #benchmark195 results
Seed 2.0 Tops Arena for Chinese Models
ByteDance's Seed 2.0 debuts at #6 text, #3 vision on LMArena, leading domestic models. Excels in math, vision perception, reasoning, and agents, matching Gemini 3 Pro. Native multimodal upgrades drive benchmark dominance.
BrowseComp-V³ Benchmark for Multimodal Agents
BrowseComp-V³ is a new benchmark with 300 challenging questions for evaluating multimodal browsing agents on deep multi-hop reasoning across text and visuals. It features subgoal-driven process evaluation and publicly searchable evidence for reproducibility. Experiments reveal state-of-the-art models achieve only 36% accuracy, highlighting integration bottlenecks.
MMDR-Bench Verifies Multimodal Research
Ohio State and Amazon release MMDR-Bench, a verifiable benchmark for multimodal Deep Research Agents. Focuses on process traceability, evidence alignment, and claim verification beyond superficial reports. Open resources include paper, GitHub, and Hugging Face datasets.
AgentLeak: Multi-Agent Privacy Leak Benchmark
AgentLeak introduces the first full-stack benchmark for privacy leakage in multi-agent LLM systems, covering internal channels like inter-agent messages. It spans 1,000 scenarios across healthcare, finance, legal, and corporate domains. Tests on top models show internal channels cause 68.9% total leakage, missed by output audits.