DeepSeek V4 Flash officially released, claims benchmark wins

DeepSeek has officially released its V4 Flash model, which the company claims outperforms its V4 Pro preview version across nine agentic benchmarks. The article verifies these claims by examining the model card and configuration files, noting that while Flash-0731 indeed wins against V4 Pro, it still falls short of Opus 4.8. The release features a 1 million token context window, utilizes a mixture-of-experts architecture with FP8 dense and FP4 expert weights, and includes an integrated speculative decoding module for improved latency and cost-efficiency in agentic applications. AI

IMPACT This release offers a cost-effective solution for agentic workloads with its optimized architecture and long context window.

RANK_REASON Frontier-lab model release with system card and technical details. [lever_c_demoted from frontier_release: ic=1 ai=1.0]

Read on dev.to — LLM tag →

• config.json
• DeepSeek
• DeepSeek Harness
• DeepSeek-V4-Flash-0731
• DSBench
• DSpark
• GGUF
• GLM-5.2
• llama.cpp
• Ollama
• Opus 4.8
• SGLang
• transformers
• Unsloth
• vLLM

• model release
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →