DeepSeek V3.5
Open SourceOpen SourceAPI Available
Benchmark Scores
89.4/100MMLU90.5
HumanEval91
Math86.5
Reasoning89
Coding90
About DeepSeek V3.5
DeepSeek V3.5 improves on V3 with better general knowledge, instruction following, and coding. It remains one of the most cost-efficient models available while delivering near-frontier performance.
Strengths
- +Incredible cost efficiency
- +Strong general performance
- +Open source
- +Fast inference with MoE
Weaknesses
- -Below frontier on hardest tasks
- -Weaker creative writing
- -Limited multimodal
Pricing
Per 1M tokens
Input$0.27
Output$1.10
Quick Facts
- Context Window
- 128K
- Parameters
- 700B (MoE, 37B active)
- Release Date
- 2026-01-20
- Category
- Open Source
Compare Models
See how DeepSeek V3.5 stacks up against other models