DeepSeek Releases V4.1-Flash As An Open Model With Lower API Pricing
DeepSeek is giving away an MIT-licensed model that benchmarks above Claude Opus 5 and GPT-5.6 Sol on several tests and scores 40 on the Artificial Analysis Intelligence Index v4.3, which puts price pressure on Western labs that sell comparable performance per token.
Reporting from 1 source: GIGAZINE.
DeepSeek released DeepSeek-V4.1-Flash on September 10, 2026. The open model is a mixture-of-experts system with 552 billion total parameters, 8 billion active parameters on input and 16 billion on output, and it supports image input. It beat Claude Opus 5 and GPT-5.6 Sol on multiple benchmark tests. Inference efficiency cut KV cache tokens to 1/3.9 of DeepSeek-V4-Flash, and off-peak API pricing is half the peak rate.
DeepSeek-V4.1-Flash is distributed on Hugging Face and ModelScope under the MIT License, with a technical report published alongside it. It uses pretraining and reinforcement learning methods that differ from DeepSeek-V4.
Artificial Analysis measured a score of 40 on its Intelligence Index v4.3, which was enough to pass GPT-5.6 Luna. On AutomationBench-AA, which tracks agent performance in clerical work such as finance and HR, the model recorded the top score ahead of GPT-6 Astra. Its output speed trails Gemini 3.8 Flash and Muse Spark 1.3, and it spends more output tokens per task than Claude Fable 5.1, so it is not the cheapest option per task among Chinese models. GLM-5.3 is slightly more cost-efficient.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.