DeepSeek-V4 challenges leading AI models on MMLU benchmark

DeepSeek has released a new AI model, DeepSeek-V4, which has demonstrated superior performance on the MMLU benchmark compared to existing leading models. This release signifies a significant advancement in the capabilities of open-source AI models. AI

IMPACT Sets new SOTA on MMLU benchmark, potentially challenging established models.

RANK_REASON Frontier-lab model release with benchmark results. [lever_c_demoted from frontier_release: ic=1 ai=1.0]

Read on Mastodon — fosstodon.org →

• DeepSeek
• DeepSeek-V4

• model release
• paper

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →