DeepSeek-V4-Flash-Vision-Exp Open Sourced, Topping Multimodal Benchmarks
智东西 · wechat · 2026-08-31
DeepSeek has open-sourced the experimental multimodal model DeepSeek-V4-Flash-Vision-Exp, integrating vision capabilities into the V4-Flash architecture. The model outperforms Opus-4.8 in 5 out of 7 text agent tasks and achieves a top score of 59.3% on the DeepSWE software engineering benchmark. In multimodal evaluations, it leads on ZeroBench and Agents'LastExam. However, it lags behind Opus-4.8 by about 10% on complex data science tasks like NL2Repo and DSBench-Hard. The DeepSeek-V4 series has seen 6 updates in the past month.
More from Models
- Meta releases Muse Spark Open: 816B-param open-weight model with 1M context, natively multimodal — IanAndrewsDC · 2026-10-16
- llama.cpp adds day-0 support for Meta's 816B Muse Spark Open within 30 minutes — ngxson · 2026-10-16
- Muse Spark vs Qwen and GLM: US open models still trail on agentic tasks — zsakib_ · 2026-10-16
- Muse Spark 816B gets llama.cpp support just 30 minutes after release — ngxson · 2026-10-16
- 24 AI Systems Recommend $314 Per Visit in Care vs Physicians' $97, a 3.2x Gap — dvir_a · 2026-10-16
- Are the fastest GPU kernels now written by AI? Kernels are hyper-verifiable, argues one dev — fleetwood___ · 2026-10-16