High Reasoning

DeepSeek v4 Flash

DeepSeek V4 Flash is a powerful 284B-parameter Mixture-of-Experts (MoE) AI model with just 13B activated parameters, delivering top-tier reasoning, coding, and knowledge performance at blazing speeds. Featuring a massive 1-million-token context window and ultra-low costs of $0.14/$0.28 per million input/output tokens, it's the efficient powerhouse for demanding tasks without breaking the bank.

1M Context

High Intelligence

No specific cutoff found Knowledge

Available for Chat, Vision, and File Uploads.

Performance Benchmarks

MMLU-Pro

86.2%

HumanEval

69.5%

GSM8K

90.8%

How do you want to interact?

Start a Conversation

Ask anything.
Have a natural conversation, brainstorm ideas, draft emails, or ask for advice.

Start Chatting

Use a Persona

Specialized Experts.
Instruct the AI to act as a Coding Tutor, Marketing Expert, or Travel Guide.

Pick a Persona

Why use DeepSeek v4 Flash?

1M Token Context Window

Supports processing entire codebases or book-length documents in a single prompt with Hybrid Attention Architecture for long conversations

High-Speed Inference

Optimized for speed at 83.8 tokens per second, faster than average for similar open-weight models, with low API pricing for high-volume use

Strong Reasoning & Coding

Excels in benchmarks like GPQA (88.1 Pass@1 Max), LiveCodeBench (91.6 Pass@1 Max), and Codeforces (3052 rating Max), approaching Pro variant performance

Capability Examples

Long Context Reasoning

Summarize this 500k-token novel excerpt while identifying key plot twists and themes.

The novel follows protagonist Alex's journey through a dystopian world. Key twists: betrayal by ally in chapter 12, hidden identity reveal in chapter 28. Themes: resilience, corruption of power. Concise analysis maintains full context coherence.

Efficient Coding Agent

Write a Python script for a multi-agent system that processes 1M tokens of log data, detects anomalies, and generates a report using MoE optimization.

python\nimport pandas as pd\nimport numpy as np\n# Efficient MoE-inspired anomaly detection for 1M logs\ndef detect_anomalies(logs: pd.DataFrame) -> dict:\n # Sparse attention simulation for efficiency\n z_scores = np.abs((logs['value'] - logs['value'].mean()) / logs['value'].std())\n anomalies = logs[z_scores > 3]\n return {'anomalies': len(anomalies), 'report': anomalies.describe()}\n# Handles 1M tokens scalably\n Script optimized for speed, runs in seconds on large datasets.

Multimodal Creative Generation

Generate a short sci-fi story with an accompanying image description and 10-second video concept, based on "cyberpunk city at midnight".

Story: In Neon-9's shadows, hacker Lena jacked into the grid, unraveling corporate secrets amid raining holograms. Image: Cyberpunk skyline with neon lights, flying cars, wet streets. Video: 10s pan from alley to towering spires, pulsing ads, dramatic synth score.

How to use

Go to Chat

Navigate to the "AI Chat" page.

Select Model

Ensure DeepSeek v4 Flash is selected.

Type Prompt

Ask a question or paste code.

Interact

Refine the answer by replying to the AI.

Made with ❤ by AI4Chat

Flash Sale 50% Off!

Daily Limit Reached

Upgrade to Premium

Credits Exhausted