# DeepSeek V4 Flash Vision Exp

As of October 21, 2026, DeepSeek V4 Flash Vision Exp ranks #51 of 123 models overall with 56.4 ±5.4 on the LLMBenchmark Index. DeepSeek V4 Flash Vision Exp ranks #12 of 50 in agentic, its strongest category. DeepSeek V4 Flash Vision Exp costs $0.44 per million input tokens and $1.32 per million output tokens. DeepSeek V4 Flash Vision Exp was released on August 21, 2026.

Page: https://llmbenchmark.ai/models/deepseek-v4-flash-vision-exp
Data from October 21, 2026. Every figure below names the board it was read from and the day.

- Maker: DeepSeek
- Released: August 21, 2026
- Price per million input tokens: $0.44
- Price per million output tokens: $1.32
- Context window: 1.05M tokens
- Overall rank: 51 of 123, by the LLMBenchmark Index
- LLMBenchmark Index: 56.4 ±5.4 (90% margin, from 3 of 7 inputs)
- Artificial Analysis Intelligence Index: 34.8
- Vibe Coding Index: not enough data (one index benchmark; at least two are needed to rank)

Benchmark results (14 of 195):

- LiveBench coding: 37.6% (source: LiveBench, read October 21, 2026)
- AA Intelligence Index: 34.8 (source: Artificial Analysis (via OpenRouter), read October 21, 2026)
- AA Coding Index: 65.0 (source: Artificial Analysis (via OpenRouter), read October 21, 2026)
- AA Agentic Index: 47.5 (source: Artificial Analysis (via OpenRouter), read October 21, 2026)
- LiveBench reasoning: 85.4% (source: LiveBench, read October 21, 2026)
- LiveBench data analysis: 79.5% (source: LiveBench, read October 21, 2026)
- LiveBench research: 69.2% (source: LiveBench, read October 21, 2026)
- LiveBench browser: 64.0% (source: LiveBench, read October 21, 2026)
- LiveBench long agentic: 39.0% (source: LiveBench, read October 21, 2026)
- LiveBench design: 70.7% (source: LiveBench, read October 21, 2026)
- GPQA Diamond (Artificial Analysis run): 91.3% (source: Artificial Analysis, read October 21, 2026)
- Humanity's Last Exam (Artificial Analysis run): 34.5% (source: Artificial Analysis, read October 21, 2026)
- SciCode: 49.7% (source: Artificial Analysis, read October 21, 2026)
- Long context reasoning (AA-LCR): 81.3% (source: Artificial Analysis, read October 21, 2026)