DeepSeek open-sources V4-Flash-Vision-Exp, its first V4-family multimodal model

DeepSeek released DeepSeek-V4-Flash-Vision-Exp weights on Hugging Face under MIT: its first experimental multimodal model in the V4 family, built on the DeepSeek-V4-Flash architecture with added visual modules and continued training. The model card benchmarks it against V4-Flash-0731 and Opus-4.8: Terminal Bench 2.1 83.9 (V4-Flash 82.7, Opus-4.8 85.0), DeepSWE 59.3 (54.4 / 58.0), and a jump on multimodal agent benchmarks — ApexBench 36.5 vs 26.2 for the text-only sibling (which ignores multimodal input). The API had been live on the DeepSeek platform since ~8/21 (set model to deepseek-v4-flash-vision-exp); the 8/31 release adds the weights and a minimal PyTorch reference implementation.

This is the Chinese open-weight wave extending from text-only coding agents into multimodal agents: open MIT weights that hold near-frontier text-agent numbers while adding vision, at Flash-tier serving cost. Teams already self-hosting V4-Flash get a drop-in path to screenshot/OCR/chart agent workflows without switching model families — though the -Exp label, vendor-only benchmarks, and early vLLM friction mean it is a preview to evaluate, not a default to deploy.