DeepSeek announced yesterday (July 31) through an API documentation release log that the official DeepSeek-V4-Flash API had entered public beta. AI benchmarking platforms such as @ArtificialAnlys and @arena also updated their relevant benchmark results today (August 1).
@ArtificialAnlys (Artificial Analysis) is a leading international independent AI benchmarking and analysis organization specializing in unbiased performance and hosting-cost evaluations across multiple fields, including large language models (LLMs), video generation, and AI music.
The Artificial Analysis Intelligence Index (AII) is its core comprehensive technical benchmark, used to measure the absolute intelligence levels of AI models and track technological evolution.
The report stated that DeepSeek V4 Flash 0731 scored 50 points on the Artificial Analysis Intelligence Index (AII), 10 points higher than DeepSeek V4 Flash (released in April 2026) and 6 points higher than DeepSeek V4 Pro.
DeepSeek V4 Flash 0731's Intelligence Index score was 1 point lower than GPT-5.6 Luna's highest score of 51 points. Even after OpenAI cut the price of GPT-5.6 Luna by 80% today, DeepSeek V4 Flash 0731's per-task cost on its own API remains approximately 60% lower than GPT-5.6 Luna's, which has a highest score of 51 points.
The key factor behind this difference is that DeepSeek provides an approximately 98% cache-hit discount for its own API, far exceeding the 90% cache-hit discount offered by most providers in the industry.
@arena (Arena.ai) is currently the global AI industry's most authoritative platform for real-time head-to-head battles and evaluations of large models based on human preferences. It originated from the renowned LMSYS Chatbot Arena team and uses an Elo rating system similar to that of chess, ranking AI models worldwide through rigorous blind human developer testing (Blind A/B Testing).
Among its specialized leaderboards, Frontend Code Arena (also known as WebDev Arena) is currently the gold standard and an industry bellwether for evaluating large models' web development, frontend engineering, and agent-level automated coding capabilities.
The report stated that DeepSeek-V4-Flash-High reshaped the Frontend Code Arena leaderboard with a score of 1586 points.
The price per MToken is $0.14 (Note: approximately RMB 0.95 at the current exchange rate) / $0.28 (approximately RMB 1.9 at the current exchange rate), making it the most cost-effective product of its kind.
In the frontend coding field, it ranks 7th overall and 3rd in the Open category! Across the various categories, it ranks 4th in Consumer Products, 6th in Reference-Based Design, Data & Analytics, and Games, and 7th in Branding & Marketing.
Compared with DeepSeek-V4-Flash-High-Preview, this represents a significant improvement, with performance increasing by 154 points. Compared with DeepSeek-V4-Pro-Preview, the improvement is even greater, reaching 121 points.
ITHome searched the X platform and found that many developers had shared comparison videos using the official version of DeepSeek-V4-Flash: