DeepSeek Model Ranks No. 12 on Web Development Coding Board

DeepSeek-V4.1-Flash-1017 ranked No. 12 in the Oct. 18 snapshot of a coding leaderboard, taking the highest open-weight position in its 146-model table. The board tests web development and coding agents that work through multiple steps and use tools.

DeepSeek-V4.1-Flash-20261017 (Max) carried an MIT license label. It scored 1,670 and received 1,417 votes. All 11 models placed above it carried proprietary labels. Kimi K3 (Max) ranked No. 15 with a score of 1,655.

DeepSeek announced its 1017 update on Oct. 17 and made the model weights available for download. The release also brought coding benchmark and developer API changes detailed in an earlier account of the update.

DeepSeek’s peak API prices are $0.30 per 1 million cache-miss input tokens and $1.20 per 1 million output tokens. Kimi K3’s API rates are $3 per 1 million input tokens and $15 per 1 million output tokens. DeepSeek’s rates equal 10% of Kimi’s for cache-miss input and 8% for output. A user’s bill also depends on token volume and how much input is served from cache.

The open-source-only table remained dated Oct. 16; license labels in the newer full table place DeepSeek first among open-weight models on Oct. 18. The ranking measures results on its selected web development and coding tasks; it does not measure developer adoption or performance across every type of coding work.