Summary

DeepSeek released V4.1-Flash, the smallest model of a new architecture family with native multimodal vision: a 552B backbone (763B total safetensors params including a 196B Engram conditional-memory module), 8B activated per token during prefill and 16B during decode, 1 shared + 384 routed experts with 6 active. The CED design organizes 40 layers as a 20-layer causal encoder followed by a 20-layer decoder whose KV cache is projected from final encoder hidden states; CSA2 provides three static attention modes (Full/Reindex/Reuse) with a hierarchical sparse indexer, FP4 (E2M1) main KV caching at 890 bytes/token (~1/4 of V4-Flash; SWA bounded replay cuts persistent KV to ~1/8), plus Single-Pass mHC kernels and DSpark speculative decoding. Pre-trained on 45T multimodal tokens; context extended to 1M. Benchmarks (reasoning_effort=100): Terminal-Bench 2.1 90.6, TB 3.0 30.0, TB 4.0 31.2, DeepSWE v1.1 74.2 (vs V4-Flash 54.4), GPQA Diamond 90.9, Codeforces 3471, HLE-with-tools 63.9, CyberGym 88.1, ExploitGym 15.3. MIT license, ungated, day-0 vLLM/SGLang/Docker Model Runner/Transformers support (deepseek-recipe libraries; no Jinja chat template). API: the model name is now deepseek-flash ($0.30 peak / $0.15 off-peak input, $1.20 / $0.60 output per 1M tokens; cache-hit $0.006 peak); V4-Flash and V4-Flash-Vision-Exp are retired from the API with legacy names temporarily routed to V4.1-Flash; V4 Pro API service continues after 2026-09-14 with unchanged billing; reasoning effort is a controllable integer 1-100. 75,774 downloads + 1,741 likes within ~1 day; #1 on HF trending at capture.

Why it matters
This is the first same-tier answer to the GLM-5.3-Flash cohort from DeepSeek, and by Terminal-Bench 2.1 (90.6 vs GLM-5.3's 88.2) it claims the top open-weight agentic-coding spot. The architecture matters as much as the scores: encoder-decoder KV projection + FP4 KV + conditional memory attacks the serving-cost side that hybrid linear attention attacks elsewhere — and the aggressive API price ($0.30/$1.20 peak) resets the cheap tier. Watch for serving friction: the deepseek-recipe runtime requirement is a new integration cost for existing stacks.
Technical details
Release HF repo created 2026-09-10T02:17:58Z; MIT; image-text-to-text; 75,774 dl + 1,741 likes in ~1 day (captured 2026-09-11T18:34Z)
Architecture 552B backbone / 763B total safetensors (incl. 196B Engram conditional memory); 8B prefill / 16B decode activated; MoE 1 shared + 384 routed, 6 active; CED = 20-layer causal encoder + 20-layer decoder (KV projected from encoder final hidden states); CSA2 static attention modes Full/Reindex/Reuse + hierarchical sparse indexer; FP4 E2M1 main KV, 890 B/token (~1/4 V4-Flash; ~1/8 with SWA bounded replay); Single-Pass mHC; DSpark speculative decoding; DeepSeek-ViT vision (2D-RoPE, 3x3 pixel-unshuffle)
Training 45T multimodal tokens pre-training; sparse attention at 64K; context extended to 1M at 34T; SFT -> RL -> on-policy distillation; reasoning_effort integer 1-100
Benchmarks TB 2.1 90.6 / TB 3.0 30.0 / TB 4.0 31.2; DeepSWE v1.1 74.2 (V4-Flash 54.4); GPQA Diamond 90.9; Codeforces 3471; HLE 36.8 (39.1 text-only); HLE w/tools 63.9; MathArena Apex 65.6; CyberGym 88.1; SEC-Bench Pro 62.8; ExploitGym 15.3; MMMU-Pro 56.5; DocVQA 95.6
API model name deepseek-flash; input $0.30 peak / $0.15 off-peak, output $1.20 / $0.60 per 1M; cache-hit $0.006 peak; concurrency 2500; max output 384K; V4-Flash + V4-Flash-Vision-Exp retired (legacy names temporarily routed); V4 Pro continues past 2026-09-14 unchanged billing
Serving day-0 vLLM / SGLang / Docker Model Runner / Transformers; deepseek-recipe Rust libs + Python reference; no Jinja chat template
Updates
2026-09-13 2026-09-13 run, serving-ecosystem check: vLLM now carries 63 issues/PRs mentioning V4.1-Flash (13 on 2026-09-12 alone - encoder CUDA-graph support #56625 17:54Z, mHC post-block folding into the delayed pre-projection #56633, ROCm gfx950 re-enable #56638; ROCm performance RFC #56506 opened 9/11); SGLang shows 32 matches (MI350X/gfx950 support #39186, Mooncake Engram KV backend with CUDA Graph #39205, prefill context parallelism #39179) plus a third-party report that fp8 wo_a absorb GEMM output is silently ~25% wrong when DEEPGEMM_SCALE_UE8M0=false on SM121 (#39193, 06:21Z). The HF model card still ships no serving recipe (community request in #13); deepseek-ai/deepseek-recipe is a prompt-format conversion library, not serving. 15+ third-party deployment repos created/updated on 9/12 target DGX Spark/GB10, Mac Studio 512GB and EXL3 quantizations; LiteLLM pricing PRs #40919/#40920 (9/12 23:36Z) add V4.1-Flash across five providers. Downloads @2026-09-13T00:03Z: 140,636 (+85.6% vs ~75.8k @9/11, ~2.5 days after release) + 2,005 likes. No third-party Terminal-Bench/SWE run yet. Stays TRIAL.
2026-09-14 2026-09-14 report run: vLLM main merged DeepSelect sparse-indexer TopK (#56464; 191 us vs 616 us on GB200 at batch 256 / 1M KV, kernel-only) and production paths for Engram DP sharding, async CPU-offload prefetch and shared-memory tables (#56512; two-node TP1xDP8 validation, <=0.9% reported TTFT gain). Not yet tagged. Recommendation stays TRIAL.
Tags
foundation-modelopen-weightsdeepseekmultimodal1m-contextkv-cachemoeapi-pricing