DeepSeek V4.1 Flash 1017

Faster, leaner agentic work on DeepSeek's KV-compressed million-token model

Try now

Read Docs

About model

DeepSeek V4.1 Flash 1017 is an update to DeepSeek V4.1 Flash, improving reasoning, coding, agentic, and visual performance while cutting task execution time and output length, so agent tasks finish faster and at lower cost. It keeps the V4.1 Flash architecture built around compressing the KV cache: a Causal Encoder-Decoder design activates just 8B of its 552B backbone parameters during prefill and 16B during decode, cutting cost on input-heavy agentic workloads, while Compressed Sparse Attention 2 and FP4 KV caching hold the global cache to 890 bytes per token across a 1M-token context. The model processes images and text natively, and reasoning effort is continuously adjustable from 1 to 100 per request. Released under the MIT license. Available on Together AI.

‍

Backbone Parameters

552B

8B activated in prefill and 16B in decode, plus 196B of Engram lookup memory

Bytes of KV Cache per Token

890

Built for cheap million-token context on long agent runs

Reasoning Effort Range

1-100

Continuously adjustable per request, trading inference cost for accuracy

Model key capabilities

Faster Agent Runs: Shorter outputs and reduced task execution time versus the original V4.1 Flash, so agent tasks complete faster and cost less
Agentic Coding: Gains across terminal, repository-scale, and program-generation tasks over the original V4.1 Flash
Input-Heavy Efficiency: Encoder-decoder design activates 8B parameters in prefill and 16B in decode, with a compressed KV cache for million-token sessions
Production-Ready Infrastructure: 99.9% SLA, available on serverless and dedicated infrastructure

Performance benchmarks

Model	FrontierMath Tier 4	GPQA Diamond	HLE	SciCode	GDPval-AA	Terminal-Bench 2.1	Agent Arena	FrontierCode	DeepSWE
DeepSeek V4.1 Flash 1017			42.6						77.0
Related open-source models
Competitor closed-source models
Claude Fable 5	87.8%	92.6%	53%	60%	62%	85%		53.5%	70%
Claude Opus 5	73.2%	93.2%	53%	56%	68%	89%	+16.4pp	53.4%	74%
GPT-5.6 Sol	82.9%	94.1%	47%	56%	61%	88%		47.5%	73%
Grok 4.5	24.4%	93.1%	40%	54%	51%	82%	+4.2pp	42.4%	54%
GPT-5.6 Luna	61.0%	91.1%	37%	53%	54%	81%		39.8%	67%

API usage

cURL
Python
Typescript

Endpoint:

deepseek-ai/DeepSeek-V4.1-Flash-1017

curl -X POST "https://api.together.xyz/v1/chat/completions" \
  -H "Authorization: Bearer $TOGETHER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-ai/DeepSeek-V4.1-Flash-1017",
    "messages": [
      {
        "role": "user",
        "content": "What are some fun things to do in New York?"
      }
    ]
}'

from together import Together

client = Together()

response = client.chat.completions.create(
  model="deepseek-ai/DeepSeek-V4.1-Flash-1017",
  messages=[
    {
      "role": "user",
      "content": "What are some fun things to do in New York?"
    }
  ]
)
print(response.choices[0].message.content)

import Together from 'together-ai';
const together = new Together();

const completion = await together.chat.completions.create({
  model: 'deepseek-ai/DeepSeek-V4.1-Flash-1017',
  messages: [
    {
      role: 'user',
      content: 'What are some fun things to do in New York?'
     }
  ],
});

console.log(completion.choices[0].message.content);

Model card
Architecture Overview:
• Same architecture as DeepSeek V4.1 Flash: a 40-layer Causal Encoder-Decoder (20-layer causal encoder, 20-layer decoder) with the decoder's global KV cache projected from the final encoder hidden states
• 552B backbone parameters activating 8B per token in prefill and 16B in decode, plus 196B parameters of sparsely accessed Engram conditional memory
• Compressed Sparse Attention 2 shares KV and indices across layers; with FP4 KV caching, the global cache stays at 890 bytes per token
• Single-token decode FLOPs rise by only about a quarter as context grows 256-fold from 4K to 1M tokens
• Native image input through a vision encoder trained from scratch; DSpark speculative decoding

Training Methodology:
• An update to DeepSeek V4.1 Flash targeting reasoning, coding, agentic, and visual performance
• DeepSeek reports reduced task execution time and output length versus the original release across multiple agent harnesses

Performance Characteristics:
• Humanity's Last Exam (no tools): 42.6, up from 36.8; text-only subset 44.0, up from 39.1
• Terminal coding (Terminal-Bench 4.0): 42.3, up from 31.2
• Repository-scale engineering (DeepSWE v1.1): 77.0, up from 74.2; repository generation (NL2Repo-Bench): 68.2, up from 65.4; program generation (ProgramBench): 29.1, up from 20.3
• Agent tasks (AutomationBench v1.0.6: 60.5, up from 54.8; Agents' Last Exam: 35.4, up from 31.8; Apex Bench v1.1: 72.1, up from 64.6)
• Vision with tools (Chartography: 83.0, up from 78.9; BabyVision: 92.5, up from 89.6; ZeroBench-main: 51.0, up from 49.0)
• All baselines are the original DeepSeek V4.1 Flash
‍
Prompting
Together AI API Access:
• Access DeepSeek V4.1 Flash 1017 via Together AI APIs using the endpoint deepseek-ai/DeepSeek-V4.1-Flash-1017
• Authenticate using your Together AI API key in request headers
• Set reasoning_effort as an integer from 1 to 100 to trade inference cost for accuracy
• Recommended sampling: temperature 1.0 and top-p 1.0
• Allow generous output budgets: DeepSeek recommends max_tokens of 256K or more within the 1M-token context
• Supports text and image input in a single request
• Available on Together AI serverless and dedicated infrastructure
‍
Applications & use cases
Cost-Sensitive Agent Systems:
• Run agents whose costs are dominated by large inputs such as codebases, logs, and document sets
• Finish agent tasks faster with shorter outputs per run
• Tune reasoning effort per step, from quick tool routing to maximum-effort planning

Software Engineering:
• Resolve repository-scale engineering tasks end to end
• Run terminal-based agent sessions that complete real environment tasks
• Generate programs and full repositories from natural language specifications

Multimodal & Long-Context Analysis:
• Reason over charts, documents, and images alongside text
• Cross-reference million-token working sets in a single request
• Combine visual understanding with tool use for chart and document workflows
‍