ModelBeat
Models/DeepSeek-V4-Flash
All models
D

DeepSeek: DeepSeek-V4-Flash

deepseek/deepseek-v4-flash

DeepSeek-V4-Flash is DeepSeek’s open-weights AI model, released Apr 24, 2026. It costs $0.05/M input and $0.14/M output tokens, and handles a 1M-token context window. Its strongest use case is agentic & tools, where it ranks in the 78th percentile of the models Model Beat tracks.

DeepSeek V4 Flash is an efficiency-optimized Mixture-of-Experts model from DeepSeek with 284B total parameters and 13B activated parameters, supporting a 1M-token context window.

Context
1M
Input
$0.05 / 1M
Output
$0.14 / 1M
Released
Apr 24, 2026
Modalities
texttext
Strongest atAgentic & Tools top 22%

Benchmarks

Scores on standardized evaluations. Higher is better — and rank shows where DeepSeek-V4-Flash lands among all models tracked on Model Beat.

49.0
Intelligence Index
Epoch AI
49th percentile of tracked models
33.0
Coding Index
Epoch AI
33rd percentile of tracked models
78.0
Agentic Index
Epoch AI
78th percentile of tracked models

Reasoning

4 evals
GPQA Diamond86.7%

Graduate-level scientific reasoning

Humanity's Last Exam30.3%

Frontier of human expert knowledge

SimpleBench46.3%

Common-sense trick questions

WeirdML45.6%

Novel ML problem-solving

Coding

2 evals
SciCode40.2%

Scientific research coding

WebDev Arena1431

Human-rated web development

Agentic & Tools

2 evals
APEX35.0%

Multi-step agentic tasks

τ²-bench95.6%

Tool-agent-user reliability

In the news

12

Changelog

15
  • DeepSeek-V4-Flash: cheapest credible provider via OpenInference now $0.14 per 1M output tokens (was $0.168, -17%)cheapest provider · Sep 18, 2026
  • DeepSeek-V4-Flash: cheapest credible provider via OpenInference now $0.05 per 1M input tokens (was $0.0679, -26%)cheapest provider · Sep 18, 2026
  • DeepSeek-V4-Flash: reported SciCode score fell from 45.3% to 40.2% (-11%)benchmark · Sep 8, 2026
  • DeepSeek-V4-Flash: reported Humanity's Last Exam score fell from 34.8% to 30.3% (-13%)benchmark · Sep 8, 2026
  • DeepSeek-V4-Flash: reported WebDev Arena score fell from 1576.54 to 1431.07 (-9%)benchmark · Aug 26, 2026
  • DeepSeek-V4-Flash: cheapest credible provider via DigitalOcean now $0.0679 per 1M input tokens (was $0.084, -19%)cheapest provider · Aug 12, 2026
  • DeepSeek-V4-Flash: cheapest credible provider via DigitalOcean now $0.084 per 1M input tokens (was $0.13, -35%)cheapest provider · Aug 7, 2026
  • DeepSeek-V4-Flash: cheapest credible provider via DigitalOcean now $0.168 per 1M output tokens (was $0.28, -40%)cheapest provider · Aug 7, 2026
  • DeepSeek-V4-Flash: Humanity's Last Exam improved from 32.1% to 34.8% (+8%)benchmark · Aug 6, 2026
  • DeepSeek-V4-Flash: cheapest credible provider via SiliconFlow now $0.28 per 1M output tokens (was $0.1932, +45%)cheapest provider · Jul 26, 2026
  • DeepSeek-V4-Flash: cheapest credible provider via SiliconFlow now $0.13 per 1M input tokens (was $0.0966, +35%)cheapest provider · Jul 26, 2026
  • DeepSeek-V4-Flash: cheapest credible provider via StreamLake now $0.1932 per 1M output tokens (was $0.154, +25%)cheapest provider · Jul 15, 2026
  • DeepSeek-V4-Flash: cheapest credible provider via StreamLake now $0.0966 per 1M input tokens (was $0.077, +25%)cheapest provider · Jul 15, 2026
  • DeepSeek-V4-Flash: cheapest credible provider via StreamLake now $0.154 per 1M output tokens (was $0.196, -21%)cheapest provider · Jul 13, 2026
  • DeepSeek-V4-Flash: cheapest credible provider via StreamLake now $0.077 per 1M input tokens (was $0.098, -21%)cheapest provider · Jul 13, 2026

Compare DeepSeek-V4-Flash with

Frequently asked questions

What is DeepSeek-V4-Flash?

DeepSeek V4 Flash is an efficiency-optimized Mixture-of-Experts model from DeepSeek with 284B total parameters and 13B activated parameters, supporting a 1M-token context window.

Who created DeepSeek-V4-Flash?

DeepSeek-V4-Flash is an AI model developed by DeepSeek, part of the DeepSeek family, released Apr 24, 2026.

How much does DeepSeek-V4-Flash cost?

DeepSeek-V4-Flash is served from $0.05 per million input tokens and $0.14 per million output tokens (cheapest credible provider on OpenRouter: OpenInference).

What is DeepSeek-V4-Flash's context window?

DeepSeek-V4-Flash supports a context window of up to 1M tokens.

How does DeepSeek-V4-Flash perform on benchmarks?

On standardized evaluations tracked by Epoch AI, DeepSeek-V4-Flash scores 95.6% on τ²-bench, 35.0% on APEX, 30.3% on Humanity's Last Exam.

Is DeepSeek-V4-Flash open source?

DeepSeek-V4-Flash is released with open weights (Open weights (unrestricted)), so the model can be downloaded and self-hosted.

Benchmark data from Epoch AI (CC BY) and Artificial Analysis; pricing, specs & descriptions from OpenRouter. Data updated Sep 18, 2026.