DeepSeek releases V4.1-Flash, says it outperforms flagship V4-Pro
Source: SiliconANGLE (opens in a new tab) · Duncan Riley
Intel Summary
Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co. Ltd. has released DeepSeek-V4.1-Flash, an open-weight model that is the smallest variant in a new architecture family. The company claims third-party evaluations demonstrate that V4.1-Flash outperforms its larger DeepSeek-V4-Pro model in performance, speed, total runtime, and cost.
Why It Matters
If validated, a smaller open-weight architecture exceeding larger flagship performance offers significant inference cost reductions and latency improvements. Teams building on DeepSeek infrastructure must evaluate V4.1-Flash ahead of stated operational changes affecting V4-Pro requests starting September 14.
Part of an ongoing development
SourceDeepSeek released V4.1-Flash
DeepSeek has released V4.1-Flash, a multimodal model containing 552 billion total parameters with 16 billion active parameters per token. According to benchmark results reported from DeepSeek, V4.1-Flash narrowly outperforms Opus 5 and GPT-5.6 Sol on the DeepSWE coding benchmark and is published under an open MIT license. Claims are as reported; this summary makes no determination about accuracy or significance.
More coverage of this development
Organizations & Entities
Topics
Related Intelligence
- ReportSame development
New Deepseek model V4.1-Flash cuts memory needs for AI agents
DeepSeek has released V4.1-Flash, a multimodal model containing 552 billion total parameters with 16 billion active parameters per token. The model reduces KV cache memory consumption to one-quarter of its predecessor. According to benchmark results reported from DeepSeek, V4.1-Flash narrowly outperforms Opus 5 and GPT-5.6 Sol on the DeepSWE coding benchmark and is published under an open MIT license.
The Decoder - DevelopmentDevelopingAlso involving DeepSeek
Intelligence agencies warn of Chinese AI model distillation
Three intelligence agencies issued a joint warning stating that Chinese companies—including DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI—have used model distillation tactics to extract billions of tokens from U.S. Claims are as reported; this summary makes no determination about accuracy or significance.
3 independent sources - DevelopmentDevelopingAlso involving China
Mistral AI raises €3B in Series D funding
Mistral AI announced that it has secured €3 billion in a Series D funding round at a post-money valuation exceeding €21 billion. Claims are as reported; this summary makes no determination about accuracy or significance.
2 independent sources - DevelopmentNewAlso involving DeepSeek
DeepSeek plans Huawei chip data center in Inner Mongolia
The Decoder reports that DeepSeek plans to establish a data center in Inner Mongolia equipped with 160,000 Huawei Ascend-950DT chips dedicated strictly to inference rather than training. If completed, the facility would represent the largest known Huawei AI processor cluster, though reported production bottlenecks mean Huawei may take over a year to deliver the hardware. Claims are as reported; this summary makes no determination about accuracy or significance.
1 reporting source