DeepSeek releases V4-Flash-Vision-Exp, an open vision model with strong coding benchmarks

zainhas · x · 2026-08-31

DeepSeek has published DeepSeek-V4-Flash-Vision-Exp on Hugging Face, an experimental image-text-to-text model under 500B parameters, released under the MIT license.\n\nPer the model card evals, it scores 83.9 on Terminal-Bench 2.1 (rank 8 in its size class) and 59.3 on Deep-SWE (rank 5 in its size class). The repo picked up 200+ likes within hours of release.

Related event: DeepSeek Quietly Open-Sources V4-Flash-Vision-Exp, Its First Multimodal Model(12 posts)→

Original post →

More from Models

Models channel →