Back
DeepSeek

DeepSeek V3.2

DeepSeek, China168/191 correctLast updated Mar 13, 2026Website

DeepSeek V3.2 by DeepSeek achieved an overall score of 86.8% on Appwrite Arena, ranking #9 of 10 benchmarked models. The benchmark tests how well AI models understand Appwrite - the open-source backend platform for authentication, databases, storage, functions, and more. This model answered 168 of 191 questions correctly across categories including Auth, Databases, Functions, Storage, and CLI. During the benchmark it consumed 2,723,075 tokens costing $0.7281, averaging 26.1 tokens per second over 106m 59s. Compare DeepSeek V3.2 with other LLMs on the Appwrite Arena leaderboard.

Total tokens
2,723,075Input: 2,555,366 Output: 167,709
Total duration
106m 59s
Avg speed
26.1 tok/s
Total cost
$ 0.7281
Cost/1MOverallFoundationAuthDatabasesFunctionsStorageSitesMessagingRealtimeCLI
$0.26Input: $0.26/1M tokens
Output: $0.38/1M tokens		86.8%94%88.8%80%88.5%93.8%94%93.5%72%80.2%
Deterministic16/17 correct
AI-Judged93% average score