DeepSeek launches first multimodal V4 model; vLLM adds native support
solyarisoftware · x · 2026-09-01
DeepSeek released its first experimental multimodal model in the V4 family, DeepSeek-V4-Flash-Vision-Exp. It adds a 32-layer vision tower to the 285B/13B V4-Flash MoE backbone, maintaining 1M context and scoring 83.5% on OCRBench. Benchmarks show a major leap in multimodal agent performance, approaching Opus-4.8, while keeping text capabilities steady. vLLM now supports it natively with a dedicated Docker image, requiring 168GB disk space and 202GB VRAM.
More from Infra
- Copper to optical: optical backplanes seen as necessary as racks move to optics — BenBajarin · 2026-10-16
- Polymarket prices 28% odds a US state enacts data center moratorium by year-end — Polymarket · 2026-10-16
- Pennsylvania residents reject $10,000-per-household offer for massive Amazon data center — Polymarket · 2026-10-16
- PyTorch now runs natively on AWS Trainium with zero code changes — PyTorch · 2026-10-16
- Chinese labs run on ~1/10 US compute; ByteDance pays 2-4x to hoover up SEA supply — robleclerc · 2026-10-16
- Zymtrace raises $12.2M and hires ML engineers to build agents that fix GPU bottlenecks — cfregly · 2026-10-16