DeepSeek-V4-Flash-High from @deepseek_ai has made significant strides in the Frontend Code Arena, scoring 1586 and ranking #7 overall and #3 among open models. This new model, which entered public beta with enhanced agent capabilities and full adaptation for the Responses API format, demonstrates an impressive performance-per-dollar ratio, priced at $0.14/$0.28 per MToken. The model showed notable improvement over its predecessors, achieving a +154 point jump from DeepSeek-V4-Flash-High-Preview and a +121 point increase from DeepSeek-V4-Pro-Preview, establishing its strong position in categories like Consumer Product and Gaming.

DeepSeek: DeepSeek is an AI research organization that builds and releases large language models optimized for coding and agentic tasks. It recently launched DeepSeek-V4-Flash-High in public beta, demonstrating major gains in frontend development benchmarks and API compatibility.
@deepseek_ai: @deepseek_ai serves as the official communication channel for the DeepSeek team on X. It announces model releases, performance updates, and technical integrations such as native support for the Responses API format and Codex.

Model Release: DeepSeek-V4-Flash-High entered public beta with upgraded agent capabilities and full adaptation for the Responses API format.
Benchmark Focus: The model targets frontend coding performance, placing among top open models in categories including Consumer Product, Reference-based Design, and Gaming.