stealthmodels.

MODEL BENCHMARKS

DeepSeek-V4.1-Flash: Selfie

Pairs a sparse reasoning backbone with 196B parameters of token-indexed Engram memory. Its encoder-decoder design sharply reduces the KV cache needed for long context.

DATASHEET

DeepSeek-V4.1-Flash specs and pricing

A good fit forAgents that repeatedly read large documents, codebases or conversation histories. Engram host-memory offload is useful when GPU memory is the constraint.

DeepSeek-V4.1-FlashOther modelsMedian

PriceDeveloper API, input / output per 1M tokens
$0.283 / $1.131Cheaper than 34 of 46 others
Context windowTokens in one request
1,000,000Larger than 7 of 49 others
Output speedTokens per second, our measurement
242 tokens/sFaster than 40 of 51 others
Developer
DeepSeek-AI
Released
10 Sep 2026
Family
DeepSeek V4.1
License
MIT
Open model
Yes
Local model
No
Origin
China
Parameters
763.21B
Active parameters, prefill
8B
Active parameters, decode
16B
Backbone parameters
551.57B
N-gram table parameters
196.61B
Max output
393,216 tokens
Input → output
Text, image → text
API model ID
deepseek-v4.1-flash
Architecture
20-layer encoder + 20-layer decoder with CSA2, Single-Pass mHC, Engram n-gram memory and DSpark/MTP

Self-host DeepSeek-V4.1-Flash

Hardware
Large multi-GPU / rack-scaleEstimated
Download size
510 GB (native mixed released precision)

Official Hugging Face weights

All prices and providers2

Alibaba Cloud Model Studio; input 0.283 USD/1M; cached read 0.03/1M; output 1.131/1M; Representative busy-period rate; documented idle-period rates are lower and region-specific.

  • OpenRouterObserved 2026-10-23
    Input $0.3Output $1.2Cached read $0.006

    per 1M tokens

    deepseek/deepseek-v4.1-flash
  • Alibaba Cloud Model StudioObserved 2026-10-06
    Input $0.283Output $1.131Cached read $0.03

    per 1M tokens

    Available

    deepseek-v4.1-flash
Deployment details
  • Exact released checkpoint tensor payload510 GB · native mixed released precision

Host RAM offload with asynchronous prefetch is supported; GPU residency depends on the serving engine and configuration.

Engram stores token-indexed vectors that can be prefetched from system RAM.

An independent NVMe implementation also exists; its reported performance has not been reproduced by us.

The released Engram tables occupy about 203GB including scales. Other checkpoint tensors occupy about 308GB, before KV cache and runtime overhead. These are storage footprints, not minimum VRAM requirements.

Engram native storage
202.76 GB
Remaining native storage
307.53 GB
Measurements and sources18

StealthMark score

StealthMark measures how well a model’s abilities carry over to original, demanding tasks it could not have trained for: physical and spatial reasoning, planning across many steps, and precise visual work. How StealthMark works explains every benchmark, what it tests and why it is hard.

General Intelligence

The ability to reason, plan, solve problems and understand how things relate and behave. A high score means it thinks hard problems through, from careful analysis to plans that span many steps.

Visual Intelligence

The ability to understand visual intent and turn ideas, descriptions and plans into coherent, effective visuals. A high score means design ideas come out the way you meant them, in illustrations, icons, diagrams or interface visuals.

Overall score

Overall shows how well a model’s abilities carry over to original, demanding tasks, weighting every benchmark into one score. Estimated means the total includes an estimate for one missing scene.

Judging

Scores reflect how well each result meets the benchmark requirements. Game scores come from the recorded play: the points earned on each level and how each level ended. How StealthMark keeps scores honest.

Games

Each game is a sequence of levels, and earlier moves carry into later levels. A level that was not played is shown as not played and never counts as zero. Play every level of Latch or Foresight yourself to get a score calculated like the models’.

Reasoning efficiency

Good results with little thinking score high. On each task, a model’s result and reasoning tokens are compared with every other model and combined with the Balanced Integration Score (Liesefeld & Janczyk, 2019); 50 is average. Single step shows how much reasoning a model needs to understand and solve one problem, multi-step how efficiently it keeps working on a task by building on its earlier reasoning, as in agentic workflows. Not part of StealthMark.

Cost efficiency

How much StealthMark a model delivers for its money, in StealthMark points per dollar: its overall score divided by what one benchmark task costs it on average. The cost is what our runs were charged, or their tokens at the model’s list price; benchmarks a model has not run yet are estimated from its other costs. Like StealthMark, it is absolute and compares with no other model. Not part of StealthMark.

BENCHMARK RESULTS

Compare results

This link opens the same results in the same order.

StealthMark benchmark results

Reset benchmark preferences?

Return to the default view, clear comparison selections and restore the advertisement.