Unverified50% confidenceBenchmarkExact time
在AIME-2024基准上，DeepSeek-V4 Flash的pass@1从0.39提升至0.49
1
Sources
50%
Confidence
Long-term
Relevance
7/13/2026
First Seen
Sources
DeepSeek-V4 Flash RL Training Successfully Ported to AMD MI355X GPU: A Complete Breakdown
twitterlmsysorg7/10/2026
Related Claims
Unverified盘古2.0 Flash在LiveCodeBench算法推理测试中得分85.1，仅次于DeepSeek V4 Flash的91.6，超过千问3.6 27B的83.971% similarUnverifiedDeepSeek V4 Flash开启MTP后代码生成速度从31.15 Token/s提升到37.3 Token/s，提升幅度约20%69% similarUnverifiedDeepSeek V4 Flash's KV Cache VRAM has been reduced to 7% of the V3.2 version (a 13.7x reduction).68% similarUnverifiedDeepSeek-V4 Pro has 1.6T parameters with 49B activated per inference; the Flash version has 284B parameters with 13B activated.67% similarVerifiedDeepSeek V4 Flash has 284B total parameters, activates 13B per inference, and requires approximately 150GB of VRAM.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/497622API
curl https://kongchang.com/api/v1/knowledge/claims/497622MCP
get_claim(id=497622)