DeepSeek V4-Flash 0731
The latest open-weight MoE model in DeepSeek’s V4 family — 284B total parameters, 13B active, 1M-token context. Matches Gemini 3.6 Flash and performs in the same class as GLM-5.2, with a major upgrade for agent use cases.
Architecture
MoE
284B total / 13B active
Context Window
1M
Tokens
Intelligence
50
AA Index v4.0
License
Open
Open Weights
The Latest V4-Flash, Tuned for Agents
DeepSeek V4-Flash 0731 is the latest release in DeepSeek’s V4 family and delivers a major upgrade for agent use cases. With 284B total parameters, 13B activated per token, and a 1M-token context window, it combines efficient inference with substantially stronger results than V4-Flash Preview and V4-Pro Preview.
Best for codebase engineering, cybersecurity work, and multi-tool agent workflows. Available in FP4 and FP8 quantizations. Released 31 Jul 2026.
Compact V4 MoE
Built for everyday agentic & long-context work
General Reasoning
Strong everyday reasoning across chat, Q&A, and structured generation tasks.
Document-Scale Analysis
1M-token context window handles long documents, codebases, and research corpora in a single call.
Agentic Workflows
Reliable tool use and instruction following for agents that need broad capability across diverse tasks.
Structured Generation
Strong format adherence for JSON, code, and structured outputs in production pipelines.
Strong reasoning at a fraction of the cost
Artificial Analysis Intelligence Index v4.0 scores. V4-Flash compresses the V4 family's strengths into a 13B-active MoE while retaining the 1M-token context window.
Intelligence Index
Better than 91% of models
GPQA Diamond
Better than 90% of models
τ²-Bench Telecom
Better than 95% of models
| Category | Benchmark | Score | Description |
|---|---|---|---|
| Reasoning | GPQA Diamond | 87% | Graduate-level scientific reasoning |
| Reasoning | Humanity's Last Exam | 28% | Humanity's Last Exam |
| Reasoning | τ²-Bench Telecom | 96% | AI agents in dual-control scenarios |
| Reasoning | AA-LCR | 63% | Long context reasoning evaluation |
| Reasoning | IFBench | 73% | Instruction-following accuracy |
| Reasoning | GDPval-AA | 46% | Agentic performance on real-world work tasks |
| Coding | SciCode | 42% | Python for scientific computing |
| Coding | Terminal-Bench Hard | 39% | Agentic coding & terminal use |
| Knowledge | AA-Omniscience Accuracy | 36% | Proportion of correctly answered questions |
Metrics sourced from Artificial Analysis and DeepSeek's published evaluations.
Flexible Pricing Tiers
Choose the optimal balance of speed and cost for your workflow. Prices are per 1M tokens.
| Tier | Input / 1M tokens | Output / 1M tokens |
|---|---|---|
| Overnight (24H) | $0.05 | $0.09 |
| Async | $0.07 | $0.14 |
| Realtime | $0.09 | $0.18 |
Context window natively supported up to 1M tokens.
Start Building in Minutes
DeepSeek V4-Flash 0731 is accessible via OpenAI-compatible endpoints.
from openai import OpenAI
client = OpenAI(
api_key="your-api-key-here",
base_url="https://api.doubleword.ai/v1"
)
# Long-context reasoning with DeepSeek V4-Flash 0731 (1M tokens)
response = client.chat.completions.create(
model="deepseek-ai/DeepSeek-V4-Flash-0731",
messages=[
{"role": "user", "content": "Analyze this document and extract key insights."}
],
)
print(response.choices[0].message.content)💡 Pro Tip
V4-Flash's 1M-token context shines on document-scale workloads — feed entire codebases, contracts, or research corpora in a single call. Use the Async or Overnight tier to slash costs further on bulk pipelines.