Open SourceDeepSeekReleased 2025-12

DeepSeek V4

DeepSeek's latest Mixture-of-Experts model - frontier performance at a fraction of US model costs

Context window

128K

Input / 1M tokens

Free

Output / 1M tokens

Free

Provider

DeepSeek

Open-source weights available. API pricing significantly cheaper than US frontier models.

DeepSeek V4 is a Mixture-of-Experts (MoE) architecture that achieves frontier-level performance on coding and reasoning tasks. The sparse activation design activates only a subset of parameters per token, keeping latency low while maintaining strong output quality. Available as open-source weights. Includes a Flash variant optimised for low-latency use cases.

Strengths

  • Mixture-of-Experts architecture - high capability, low activation cost
  • Open-source weights freely available
  • Strong coding and reasoning benchmarks
  • Flash variant offers low-latency inference
  • Significantly cheaper to run than US frontier models

Best for developers who...

Self-hosted deployments needing frontier performanceCost-sensitive high-volume inferenceCoding and technical tasksResearchers studying MoE architectures

Compare DeepSeek V4 with

All model comparisons →

Learn the concepts