Text

DeepSeek V3.1

Fast, general-purpose MoE model with long-context variants and strong coding ability.

Overview
V3.1 is the default model for chat and production workloads. It balances speed and quality, supports long-context workflows, and is widely used for general-purpose automation and code.
Best for: General chat, Code generation, Long documents
  • MoE architecture with fast default latency.
  • Reported long-context support in specialized variants.
  • Open-source ecosystem with broad community adoption.
Pricing
Transparent pricing for legacy models. V4 pricing will be announced at launch.
Tokens$1.00 / 1M tokens
Full pricing
Research summary
Compiled from public research notes and internal summaries. Specifications may evolve ahead of official releases.

DeepSeek V3.1 is a large, open-source MoE model built for general chat, coding, and long-context workflows. Reports describe ~685B total parameters with ~37B activated per token, which keeps inference efficient while preserving capacity.

The model is widely referenced as supporting 128K-class context in long-context variants and is trained on very large corpora (public reports cite ~14.8T tokens). The MIT license enables commercial use and internal fine-tuning, which has accelerated adoption across teams that need flexible deployment.

V3.1 is a strong default when you want balanced quality, speed, and cost. It handles summarization, extraction, and code generation well, and can anchor production systems while you route specialized tasks to R1, Math-7B, or multimodal models.

Focus areas
The traits to evaluate when choosing this model.
  • General chat, code, and automation workloads.
  • Long-context variants for large documents.
  • Sparse MoE efficiency at scale.
  • MIT-licensed for commercial deployment.
  • Stable production behavior and tuning headroom.
Validate benchmarks and latency on your own prompts before committing a production rollout.