DeepSeek logo

DeepSeek-V4-Flash-0731

DeepSeek

DeepSeek-V4-Flash-0731 is an updated fast MoE checkpoint from DeepSeek released in July 2026, focused on agentic and terminal-based tasks. It demonstrates strong performance on agent benchmarks such as Terminal-Bench 2.1, CyberGym and Toolathlon. Weights and a technical report are publicly available.

Key Specifications

Parameters
304.0B
Context
1.0M
Release Date
July 31, 2026
Average Score
74.8%

Timeline

Key dates in the model's history
Announcement
July 31, 2026
Last Update
August 27, 2026

Technical Specifications

Parameters
304.0B
Training Tokens
-
Knowledge Cutoff
-
Family
-
Capabilities
MultimodalZeroEval

Pricing & Availability

Input (per 1M tokens)
$0.14
Output (per 1M tokens)
$0.28
Max Input Tokens
1.0M
Max Output Tokens
393.2K
Supported Features
Function CallingStructured OutputCode ExecutionWeb SearchBatch InferenceFine-tuning

Benchmark Results

Model performance metrics across various tests and benchmarks

Other Tests

Specialized benchmarks
Terminal-Bench 2.1
DeepSeek Harness minimal mode; max reasoning effort; temperature 1.0; top_p 0.95Self-reported
83.0%
CyberGym
DeepSeek Harness minimal mode; max reasoning effort; temperature 1.0; top_p 0.95Self-reported
77.0%
Toolathlon
Verified split; DeepSeek Harness minimal mode; max reasoning effort; temperature 1.0; top_p 0.95Self-reported
70.0%
DSBench-FullStack
Internal full-stack development test set; DeepSeek Harness minimal mode; max reasoning effort; temperature 1.0; top_p 0.95Self-reported
69.0%

License & Metadata

License
mit
Announcement Date
July 31, 2026
Last Updated
August 27, 2026

Compare DeepSeek-V4-Flash-0731

All comparisons

Similar Models

All Models

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.