DeepSeek

DeepSeek V4 Flash 0731

Tool callingStructured outputReasoning
DeepSeek V4 Flash 0731 model card cover

Official DeepSeek V4 Flash release with enhanced agentic capabilities and integrated DSpark speculative decoding

Context
1.0M
Released
2mo ago
Parameters
—
Input / 1M tokens
$0.02
Output / 1M tokens
$0.64
Cached input / 1M tokens
$0.02

Sources:models.devOpenRouter

Benchmarks

DeepSWE is the primary coding benchmark where an exact model evaluation exists. Other sourced evaluations appear below; missing scores are not zero.

Not evaluated in DeepSWE 1.1 as of 2026-09-22. Results are only shown for exact model versions in the official dataset. Official leaderboard.

Other published benchmarks

EvaluationReported score and source
ARC-AGI-2
56.0%
Third-party evaluator · deepseek-v4-flash-0731_high
Reasoning effort: high
External evaluation via Epoch AI
Original-source terms apply
ARC-AGI-2
46.0%
Third-party evaluator · deepseek-v4-flash-0731_low
Reasoning effort: low
External evaluation via Epoch AI
Original-source terms apply
ARC-AGI-2
61.4%
Third-party evaluator · deepseek-v4-flash-0731_max
Reasoning effort: max
External evaluation via Epoch AI
Original-source terms apply
ARC-AGI-2
2.1%
Third-party evaluator · deepseek-v4-flash-0731_none
Reasoning effort: none
External evaluation via Epoch AI
Original-source terms apply
FrontierMath-Tier-4-v2-Private
24.4%
Epoch AI · deepseek-v4-flash-0731_max
Reasoning effort: max
Epoch AI evaluation
FrontierMath-Tiers-1-3-v2-Private
57.5%
Epoch AI · deepseek-v4-flash-0731_max
Reasoning effort: max
Epoch AI evaluation
GPQA diamond
91.0%
Epoch AI · deepseek-v4-flash-0731_max
Reasoning effort: max
Epoch AI evaluation
OTIS Mock AIME 2024-2025
94.4%
Epoch AI · deepseek-v4-flash-0731_max
Reasoning effort: max
Epoch AI evaluation
SimpleQA Verified
33.6%
Epoch AI · deepseek-v4-flash-0731_max
Reasoning effort: max
Epoch AI evaluation

Benchmark compilation: Epoch AI, Capabilities & benchmarking (CC BY attribution). Scores are converted to percentages and model names matched to the Vertical Tabs catalog. Individual evaluations are credited to their original sources. External results may carry separate licences and evaluation configurations; do not interpret scores from different harnesses as directly comparable.

Specifications

Developer
DeepSeek
Model family
DeepSeek
Type
Language model
Predecessor
—
Successor
—
Released
July 31, 2026
Model ID
deepseek/deepseek-v4-flash-0731
Openness
Open weights
License
MIT
Parameters
—
Input modalities
text
Output modalities
text
Modality
text->text
Tokenizer
DeepSeek
Max output tokens
944K
Knowledge cutoff
2025-05
Source freshness
Metadata synced Oct 8, 2026, 10:17 PM UTC
Additional source evidence
Source model
deepseek/deepseek-v4-flash-0731
Source family
deepseek-flash
Last updated
2026-07-31
Knowledge cutoff
2025-05
Model context
1M
Model output limit
384K
Open weights
Yes
File attachments
No
Reasoning
Yes
Tool calling
Yes
Structured output
Yes
Temperature control
Yes

Downloadable weights

Published benchmarks

BenchmarkScoreEvaluation detailsEvidence
Terminal-Bench82.7 · pass@1max · 2.1Source
NL2Repo54.2 · resolve rateDeepSeek Harness minimal mode · max effort · 2026-07-31Source
CyberGym76.7 · scoreDeepSeek Harness minimal mode · max effort · 2026-07-31Source
DeepSWE54.4 · resolve rateDeepSeek Harness minimal mode · max effort · 2026-07-31Source
Toolathlon-Verified70.3 · scoreDeepSeek Harness minimal mode · max effort · 2026-07-31Source
Agents' Last Exam25.2 · scoreDeepSeek Harness minimal mode · max effort · 2026-07-31Source
AutomationBench25.1 · success ratepublic · max effort · 2026-07-31Source
DSBench-FullStack68.7 · scoreinternal · max effort · 2026-07-31Source
DSBench-Hard59.6 · scoreinternal · max effort · 2026-07-31Source
Complete source record (JSON)
{
  "id": "deepseek/deepseek-v4-flash-0731",
  "name": "DeepSeek V4 Flash 0731",
  "description": "Official DeepSeek V4 Flash release with enhanced agentic capabilities and integrated DSpark speculative decoding",
  "family": "deepseek-flash",
  "last_updated": "2026-07-31",
  "knowledge": "2025-05",
  "attachment": false,
  "reasoning": true,
  "tool_call": true,
  "structured_output": true,
  "temperature": true,
  "modalities": {
    "input": [
      "text"
    ],
    "output": [
      "text"
    ]
  },
  "limit": {
    "context": 1000000,
    "output": 384000
  },
  "benchmarks": [
    {
      "name": "Terminal-Bench",
      "score": 82.7,
      "source": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731",
      "metric": "pass@1",
      "variant": "max",
      "version": "2.1"
    },
    {
      "name": "NL2Repo",
      "score": 54.2,
      "source": "https://api-docs.deepseek.com/updates/",
      "metric": "resolve rate",
      "harness": "DeepSeek Harness minimal mode",
      "variant": "max effort",
      "date": "2026-07-31"
    },
    {
      "name": "CyberGym",
      "score": 76.7,
      "source": "https://api-docs.deepseek.com/updates/",
      "metric": "score",
      "harness": "DeepSeek Harness minimal mode",
      "variant": "max effort",
      "date": "2026-07-31"
    },
    {
      "name": "DeepSWE",
      "score": 54.4,
      "source": "https://api-docs.deepseek.com/updates/",
      "metric": "resolve rate",
      "harness": "DeepSeek Harness minimal mode",
      "variant": "max effort",
      "date": "2026-07-31"
    },
    {
      "name": "Toolathlon-Verified",
      "score": 70.3,
      "source": "https://api-docs.deepseek.com/updates/",
      "metric": "score",
      "harness": "DeepSeek Harness minimal mode",
      "variant": "max effort",
      "date": "2026-07-31"
    },
    {
      "name": "Agents' Last Exam",
      "score": 25.2,
      "source": "https://api-docs.deepseek.com/updates/",
      "metric": "score",
      "harness": "DeepSeek Harness minimal mode",
      "variant": "max effort",
      "date": "2026-07-31"
    },
    {
      "name": "AutomationBench",
      "score": 25.1,
      "source": "https://api-docs.deepseek.com/updates/",
      "metric": "success rate",
      "variant": "max effort",
      "dataset": "public",
      "date": "2026-07-31"
    },
    {
      "name": "DSBench-FullStack",
      "score": 68.7,
      "source": "https://api-docs.deepseek.com/updates/",
      "metric": "score",
      "variant": "max effort",
      "dataset": "internal",
      "date": "2026-07-31"
    },
    {
      "name": "DSBench-Hard",
      "score": 59.6,
      "source": "https://api-docs.deepseek.com/updates/",
      "metric": "score",
      "variant": "max effort",
      "dataset": "internal",
      "date": "2026-07-31"
    }
  ],
  "release_date": "2026-07-31",
  "open_weights": true,
  "license": "MIT",
  "weights": [
    {
      "label": "Hugging Face",
      "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731"
    }
  ]
}

Infrastructure

Provider availability

All providers
ProviderEndpointContextInput /M tokensOutput /M tokensCached input /M tokensCache write /M tokensUptime 1dData handling
OpenInferenceopen-inference/fp41.0M$0.01$0.58$0.01—96.25%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Relacerelace/fp41.0M$0.02$0.64$0.02—100.00%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Sail Researchsail-research/us1.0M$0.02$0.42$0.01—100.00%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Rekareka262K$0.02$0.53$0.0056—100.00%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

DeepInfradeepinfra/fp81.0M$0.06$0.18$0.02—100.00%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Sail Researchsail-research/fp41.0M$0.10$0.30$0.03—100.00%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

DigitalOceandigitalocean1.0M$0.12$0.24$0.02—100.00%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Inceptroninceptron/fp41.0M$0.13$0.60$0.03—99.86%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

BaseTenbaseten/fp81.0M$0.13$0.26$0.03—99.30%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

BaseTenbaseten/fp8/us1.0M$0.13$0.26$0.03—98.89%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

CoreWeavecoreweave/fp8262K$0.13$0.28$0.07—100.00%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Venicevenice1M$0.13$0.26$0.03—100.00%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Waferwafer/fast1.0M$0.13$0.23$0.13—100.00%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

StreamLakestreamlake/fp81.0M$0.13$0.40$0.0042—100.00%
ZDR: No

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Coherecohere1.0M$0.14$0.28$0.07—99.31%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Parasailparasail/fp81.0M$0.14$0.28$0.05—100.00%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Togethertogether1.0M$0.14$0.28$0.03—93.81%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Mancer 2mancer/fp81.0M$0.20$0.60——99.49%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

SiliconFlowsiliconflow/fp81.0M$0.22$0.66$0.03—100.00%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

GMICloudgmicloud/fp81.0M$0.29$0.86$0.0091—100.00%
ZDR: No

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Phalaphala1.0M$0.31$0.92$0.02—100.00%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Alibabaalibaba1M$0.35$1.06$0.04—99.93%
ZDR: No

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Alibabaalibaba/us-east-11M$0.37$1.12$0.04—100.00%
ZDR: No

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Novitanovita/fp81.0M$0.41$1.23$0.03—100.00%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

AtlasCloudatlas-cloud/fp41.0M$0.44$1.32$0.03—100.00%
ZDR: No

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Baidubaidu/fp81.0M$0.44$1.32$0.01—100.00%
ZDR: No

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Cloudflarecloudflare1.0M$0.44$1.32$0.01—95.91%
ZDR: No

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Provider data:OpenRouter

Deployment graph

Can I run this?

Vertical Tabs only shows compatibility when it has structured evidence tying this exact model to hardware, runtime, quantization, memory use, and observed performance.

Browse hardware