DeepSeek-V4.1-Flash
A large mixture-of-experts language model released by DeepSeek, employing a 552B MoE architecture. It supports visual input and a one-million-token context window, and introduces the Engram memory architecture along with a compressed KV sharing mechanism, optimized for agentic use cases.
Core Facts
Pangu 2.0 Flash adopts a MoE (Mixture of Experts) architecture with a parameter scale of 92B A6B (92B total parameters, 6B activated parameters).
DeepSeek-V4.1-Flash adopts a MoE (Mixture of Experts) backbone architecture with 552B parameters.
DeepSeek-V4.1-Flash's KV cache is compressed to 890 bytes per token, only one-quarter that of V4 Flash.
Timeline (last 90 days)
In SemiAnalysis's AgentX benchmark, vLLM achieved the best-ever result for DeepSeek-V4.1-Flash using real agentic workloads.
After applying the low-bit tables on DeepSeek-V4.1-Flash and Qwen3.8-Flash-Next, perplexity rose by less than 1.2%, and downstream task performance was largely maintained.
On two existing large models, DeepSeek-V4.1-Flash and Qwen3.8-Flash-Next, the low-bit tables saved 47% to 55% of table memory.
The Engram n-gram memory module of DeepSeek-V4.1-Flash occupies 196 billion parameters.
The DeepSeek v4.1 Flash model can already run on conventional inference engines such as vLLM, and DeepInfra was initially built on vLLM as well.
The entire fine-tuning round cost about 23 RMB in DeepSeek API calls, plus about 9 Kaggle weekly free GPU hours.
Earlier tests by Artificial Analysis and Vals.ai showed that V4.1 Flash ranks among the top open-weight models but lags on SRE-type tasks and legal-agent benchmarks.
The top of terminal-bench-2.0 is mindlab-research/Macaron-V1-Venti (87.6), terminal-bench-2.1 belongs to DeepSeek-V4.1-Flash (90.6), and the base version terminal-bench is led by Ling-3.1-flash with 46.03.
DeepSeek-V4.1-Flash-1017 is a post-training checkpoint, not a new architecture, and is still based on V4.1-Flash released in September.
DeepSeek V4.1-Flash-1017 improves by 11 points on the Terminal-Bench 4.0 benchmark compared with the previous version.
40 more timeline events
All Facts (20)
Pangu 2.0 Flash adopts a MoE (Mixture of Experts) architecture with a parameter scale of 92B A6B (92B total parameters, 6B activated parameters).
80%VerifiedDeepSeek-V4.1-Flash adopts a MoE (Mixture of Experts) backbone architecture with 552B parameters.
75%VerifiedDeepSeek-V4.1-Flash's KV cache is compressed to 890 bytes per token, only one-quarter that of V4 Flash.
65%VerifiedDeepSeek released DeepSeek-V4.1-Flash, a method that significantly compresses the memory required for KV cache.
70%Partially VerifiedDeepSeek V4.1-Flash-1017 improves by 11 points on the Terminal-Bench 4.0 benchmark compared with the previous version.
75%UnverifiedThe release party claimed that the cost of DeepSeek-V4.1-Flash is only 1/40 that of Opus 5 and GPT-5.6 Sol.
70%UnverifiedDeepSeek-V4-Flash-Vision-Exp adopts a MoE architecture with approximately 305B total parameters, activating only approximately 13B per token, for an activation rate of approximately 4.6%.
60%UnverifiedIn SemiAnalysis's AgentX benchmark, vLLM achieved the best-ever result for DeepSeek-V4.1-Flash using real agentic workloads.
50%UnverifiedAfter applying the low-bit tables on DeepSeek-V4.1-Flash and Qwen3.8-Flash-Next, perplexity rose by less than 1.2%, and downstream task performance was largely maintained.
50%UnverifiedThe Engram n-gram memory module of DeepSeek-V4.1-Flash occupies 196 billion parameters.
50%UnverifiedOn two existing large models, DeepSeek-V4.1-Flash and Qwen3.8-Flash-Next, the low-bit tables saved 47% to 55% of table memory.
50%UnverifiedThe DeepSeek v4.1 Flash model can already run on conventional inference engines such as vLLM, and DeepInfra was initially built on vLLM as well.
50%UnverifiedThe entire fine-tuning round cost about 23 RMB in DeepSeek API calls, plus about 9 Kaggle weekly free GPU hours.
50%UnverifiedEarlier tests by Artificial Analysis and Vals.ai showed that V4.1 Flash ranks among the top open-weight models but lags on SRE-type tasks and legal-agent benchmarks.
50%UnverifiedThe top of terminal-bench-2.0 is mindlab-research/Macaron-V1-Venti (87.6), terminal-bench-2.1 belongs to DeepSeek-V4.1-Flash (90.6), and the base version terminal-bench is led by Ling-3.1-flash with 46.03.
50%UnverifiedDeepSeek-V4.1-Flash-1017 is a post-training checkpoint, not a new architecture, and is still based on V4.1-Flash released in September.
50%UnverifiedAmong six zeroth-order cases on DeepSeek-V4.1-Flash, only one new target token promotion occurred, along with one regression.
50%UnverifiedDeepSeek-V4.1-Flash has 763B parameters, BF16 weights are about 1.5 TB, the official FP4/FP8 checkpoint is 475.2 GiB, and coolbho3k's 3-bit EXL3 package compresses it to 309.3 GiB.
50%UnverifiedThe MoE structure of DeepSeek-V4.1-Flash has 40 layers, each layer has 384 routed experts, and each token activates only 6 experts per layer.
50%UnverifiedOne change in v0.32.0rc2 is to pin the evaluation of DeepSeek-V4.1-Flash-NVFP4 to its FP8-Engram version.
50%