Tag: #multimodal328 results
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.
ByteDance launches Doubao 2.0, a major multimodal Agent model upgrade with Seedance 2.0 video and Seedream 5.0 Lite image generation. It excels in multimodal understanding, enterprise Agents, and code reasoning. Multiple sizes (Pro, Lite, Mini) optimize for deployment latency and performance.
After 21 months of development, Doubao large model officially enters its 2.0 era. It claims the highest scores in visual benchmarks.
Doubao large model enters 2.0 era after 21 months of development. It claims the highest score in vision benchmarks.
ByteDance's Doubao large model officially advances to 2.0 after 21 months of development. It claims the highest scores in visual benchmarks.
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.
MAPLE is a modality-aware ecosystem for post-training multimodal LLMs, including MAPLE-bench, MAPO optimization, and adaptive curricula. It stratifies training by modality needs to cut variance and speed convergence. It closes uni/multi-modal gaps by 30% and converges 3x faster.
Introduces BLPO to optimize prompts for multimodal LLM-as-a-judge evaluating AI images. Overcomes context limits by converting images to text representations. Outperforms baselines on four datasets with three LLM judges.