Model Details
DeepSeek-V3.1
Quick answer: DeepSeek-V3.1 is an iterative update in the DeepSeek-V3 series, building on the V3-0324 checkpoint with continued improvements in coding, reasoning, and instruction following. It maintains the 685B MoE architecture, MIT license, and $0.27/$1.10 API pricing of the V3 line.
What DeepSeek-V3.1 Is
DeepSeek-V3.1 continues the versioned improvement trajectory of the DeepSeek-V3 family. Following V3 (December 2024) and V3-0324 (March 2025), the .1 update represents a more formal increment — a training checkpoint improvement with targeted quality gains rather than a date-stamped hotfix. It continues the trend of improving SWE-bench and hard reasoning scores within the same architectural framework.
Specifications
| Field | Value |
|---|---|
| Organization | DeepSeek |
| Total parameters | 685 billion (MoE) |
| Active parameters per token | 37 billion |
| Context window | 128,000 tokens |
| License | MIT |
| Modality | Text |
Pricing
| Input (per 1M tokens) | Output (per 1M tokens) | |
|---|---|---|
| DeepSeek API | $0.27 | $1.10 |
Related Models & Benchmarks
- Compare: DeepSeek-V3-0324 · DeepSeek-V3.2
- Benchmarks: SWE-bench Verified
Last updated 2026-06-19.
DeepSeek-V3.1 vs Alternatives
| Model | BrowseComp-ZH | LiveCodeBench | MMLU-Pro | License |
|---|---|---|---|---|
| DeepSeek-V3.1 | 49.2% | 56.4% | 83.7% | MIT |
| DeepSeek-V3.2 | 65.0% | — | — | MIT |
| DeepSeek-V3 0324 | — | — | 81.2% | MIT |
DeepSeek-V3.2 improved BrowseComp-ZH substantially (65.0% vs 49.2%). V3.1 remains MIT — use V3.2 if available.
Frequently Asked Questions
What is DeepSeek-V3.1?
DeepSeek's July 2025 model scoring 49.2% BrowseComp-ZH, 56.4% LiveCodeBench, 83.7% MMLU-Pro. MIT — open-source.Scores from Benchgen evaluations. Last updated 2026-07-24.