Models / DeepSeek
Chat
Reasoning

DeepSeek V4 Flash 0731

Efficient 1M-context model for agentic coding with adjustable reasoning effort

Try now
read docs

About model

DeepSeek V4 Flash 0731 is the official release of DeepSeek's efficiency-focused V4 Flash model, superseding the April preview with substantially stronger agentic capability. It is a Mixture-of-Experts model with 284B total parameters and 13B active per token, supporting a 1M-token context window on a hybrid attention design built to keep ultra-long-context inference cheap, and it ships with a speculative decoding module attached for faster generation. Reasoning effort is adjustable per request across low, high, and max levels, letting the same deployment serve quick responses and deep deliberation. DeepSeek reports the release outperforms the much larger V4 Pro preview on its published agentic benchmark set despite far fewer activated parameters. Released under the MIT license. Available on Together AI.

‍

Total Parameters (13B active)

284B

MoE routing that keeps per-token inference cost at small-model levels

Context Window

1M

Entire repositories or thousands of pages in a single request

Reasoning Effort Levels

3

Low, high, and max deliberation, set per request

Model key capabilities
  • Agentic Coding: Completes terminal, repository, and security-engineering tasks end to end, the focus of this release's largest improvements
  • Adjustable Reasoning Effort: Low, high, and max effort levels set per request, trading response speed against reasoning depth on the same deployment
  • Efficient Long Context: 1M-token window on a hybrid attention architecture designed to keep long-context inference fast and affordable
  • Production-Ready Infrastructure: 99.9% SLA, available on serverless and dedicated infrastructure

  • API usage

    • cURL
    • Python
    • Typescript

    Endpoint:

    deepseek-ai/DeepSeek-V4-Flash-0731

    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-Flash-0731",
    "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-Flash-0731",
  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-Flash-0731',
  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:
    • Mixture-of-Experts model with 284B total parameters and 13B activated per token, built on the DeepSeekMoE framework with Multi-Token Prediction retained
    • Hybrid attention combining Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA) for efficient processing of ultra-long sequences
    • Manifold-Constrained Hyper-Connections (mHC) strengthen residual signal propagation across layers
    • Routed expert parameters stored in FP4 precision; DSpark speculative decoding module attached to the released checkpoint
    • 1M-token context window

    Training Methodology:
    • Official 0731 release of V4 Flash, superseding the April preview with substantially enhanced agentic capabilities
    • DeepSeek reports the release outperforms the larger DeepSeek V4 Pro preview across its published agentic benchmarks despite a far smaller activated parameter count
    • Published agentic results evaluated at max reasoning effort with temperature 1.0 and top-p 0.95

    Performance Characteristics:
    • Terminal coding (Terminal Bench 2.1, completing real tasks in a command-line environment): 82.7, up from 61.8 for the preview
    • Repository generation (NL2Repo, building full repositories from natural language specs): 54.2
    • Security engineering (Cybergym): 76.7
    • Repository-scale software engineering (DeepSWE): 54.4, up from 7.3 for the preview
    • Multi-tool agent tasks (Toolathlon-Verified): 70.3; Agents' Last Exam: 25.2; AutomationBench Public: 25.1

    ‍

  • Prompting

    Together AI API Access:
    • Access DeepSeek V4 Flash 0731 via Together AI APIs using the endpoint deepseek-ai/DeepSeek-V4-Flash-0731
    • Authenticate using your Together AI API key in request headers
    • Set reasoning_effort to low, high, or max per request to control deliberation depth
    • Recommended sampling: temperature 1.0, with top-p 0.95 for agentic workloads and 1.0 otherwise
    • At high and max effort, allow generous output budgets: DeepSeek recommends up to 384K output tokens
    • Available on Together AI serverless and dedicated infrastructure

    ‍

  • Applications & use cases

    Agentic Software Engineering:
    • Run terminal-based agent sessions that complete real environment tasks end to end
    • Generate and modify entire repositories from natural language specifications
    • Apply the model to security-engineering and vulnerability-analysis workflows

    Multi-Tool Agent Workflows:
    • Orchestrate agents across many tools with reasoning effort tuned to task difficulty
    • Run quick tool-routing calls at low effort and hard planning steps at max on one deployment
    • Sustain long agent trajectories with generous output budgets at high effort

    Long-Context Analysis:
    • Hold entire codebases or large document sets in the 1M-token window
    • Cross-reference material across a session without external retrieval for mid-sized corpora
    • Keep long-running agent state in context across extended multi-turn sessions

    ‍

Related models
new
Chat
DeepSeek V4 Pro

  • Cached/1M

    $0.20

  • Input/1M

    $1.74

  • Output/1M

    $3.48
Function Calling
JSON Mode
Reasoning
Chat
Code
LLM
new
Chat
MiniMax M3

  • Cached/1M

    $0.06

  • Input/1M

    $0.30

  • Output/1M

    $1.20
Function Calling
JSON Mode
Reasoning
Vision
Chat
new
Chat
GLM-5.2

  • Cached/1M

    $0.26

  • Input/1M

    $1.40

  • Output/1M

    $4.40
Function Calling
JSON Mode
Reasoning
Chat
Code
Model specifications
Model data
  • Model provider
    DeepSeek
  • Type
    Chat
    Reasoning
  • Deployment
    Serverless
    Dedicated
  • Endpoint
    deepseek-ai/DeepSeek-V4-Flash-0731
  • Parameters
    284B
  • Activated parameters
    13B
  • Context length
    1M
  • Input price

    $0.14 / 1M tokens

    $0.03 (cached)/1M

  • Output price

    $0.28 / 1M tokens

  • Input modalities
    Text
  • Output modalities
    Text
  • Released
    July 30, 2026
  • Quantization level
    NVFP4
  • Category
    Chat
Run in Playground
Quickstart docs
Deploy model