Retour à Deepseek V3
DeepSeek's experimental V3.2 MoE checkpoint — preview of upcoming V3.2 improvements to reasoning and agentic capabilities.
164K tokensMoEMIT
Benchmarks
DeepSeek-V3.2-Exp
56.7%DeepSeek-V3.2
56.7%DeepSeek-V3.2-Exp Prompt+Thinking — rank #14
