And now let’s get it started! https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017…

deepseek-ai/DeepSeek-V4.1-Flash-1017 · Hugging Face

Source: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017

Technical Report👁️

https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017#introductionIntroduction

DeepSeek-V4.1-Flash-1017is a minor update toDeepSeek-V4.1-Flash, delivering substantial improvements across reasoning, coding, agentic tasks, and visual benchmarks. Alongside these gains, the model reduces task execution time and output length, enabling users to complete tasks faster and at lower cost and advancing the performance–cost Pareto frontier across multiple agent harnesses.

BenchmarkDS-V4.1-Flash-1017DS-V4.1-FlashK3GLM-5.3GPT-5.6 SolOpus 5HLE w/o tools42.6 (44.0†)36.8 (39.1†)43.542.0†44.556.3Terminal-Bench 4.042.331.212.637.939.951.8DeepSWE v1.177.074.267.566.973.074.0ProgramBench29.120.317.519.023.037.0NL2Repo-Bench68.265.458.058.056.875.3Automation-Bench v1.0.660.554.846.748.845.850.3Agents’ Last Exam35.431.827.628.526.728.6Apex Bench v1.172.164.647.856.654.365.8Chartography (w/tools)83.078.968.1—79.984.0BabyVision (w/tools)92.589.685.7—88.994.1ZeroBench-main (w/tools)51.049.041.0—53.052.0 † denotes the text-only subset of HLE. We use the Claude Code harness for Terminal-Bench 4.0 and the Mini-SWE-V2 harness for DeepSWE v1.1.

https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017#prompt-encodingPrompt Encoding

This release does not include a Jinja-format chat template. Theencodingfolder contains a self-contained Python reference implementation (encoding\.py) with test cases for multi-turn conversations, tool calling, thinking mode, numeric reasoning effort, mid-conversation system messages, and interleaved image content.

For production use, we additionally releasedeepseek-recipe, a set of Rust libraries with Python bindings that provides the same prompt format as a maintained, protocol-aware toolkit. It converts Messages, Chat Completions, and Responses API requests into the Conversation format, encodes them into DeepSeek V4 and V4.1 prompts or token IDs, and parses model output back into complete or streamed responses — covering thinking, tool calls, images, and generation settings. Model inference, tool execution, and HTTP transport are left to the caller.

https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017#minimal-inferenceMinimal Inference

Please refer to theinferencefolder for instructions on weight conversion and running inference locally.

Recommended sampling parameters:

ParameterValuetemperature1.0top\_p1.0context\_window1M tokensmax\_tokens≥ 256K

https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017#licenseLicense

This repository and the model weights are licensed under theMIT License.

https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017#citationCitation

@misc{deepseekai2026deepseekv41flash,
      title={DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression},
      author={DeepSeek-AI},
      year={2026},
}

https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017#contactContact

If you have any questions, please raise an issue or contact us at[email protected].

DeepSeek (@deepseek_ai): 🔓 Open-source release

Get DeepSeek-V4.1-Flash-1017: 🤗 Hugging Face: https://t.co/PxjsLqLx2m 🔗 ModelScope: https://t.co/ATub2olG2H

4/4