DeepSeek: DeepSeek V3.1 Base

Text input Text output
Author's Description

This is a base model, trained only for raw next-token prediction. Unlike instruct/chat models, it has not been fine-tuned to follow user instructions. Prompts need to be written more like training text or examples rather than simple requests (e.g., “Translate the following sentence…” instead of just “Translate this”). DeepSeek-V3.1 Base is a 671B parameter open Mixture-of-Experts (MoE) language model with 37B active parameters per forward pass and a context length of 128K tokens. Trained on 14.8T tokens using FP8 mixed precision, it achieves high training efficiency and stability, with strong performance across language, reasoning, math, and coding tasks.

Key Specifications
Cost
$$
Context
163K
Parameters
671B (Rumoured)
Released
Aug 20, 2025
Speed
Ability
Reliability
Supported Parameters

This model supports the following parameters:

Top Logprobs Stop Top P Seed Min P Frequency Penalty Max Tokens Logprobs Presence Penalty Logit Bias Temperature
Performance Summary

DeepSeek-V3.1 Base demonstrates exceptional speed, consistently ranking among the fastest models available, and offers highly competitive pricing across various benchmarks. As a base model, it is designed for raw next-token prediction rather than direct instruction following, which is evident in its benchmark performance. In coding, it achieved 58.3% accuracy, placing it in the 27th percentile, with a cost efficiency in the 91st percentile. However, its performance in instruction following and reasoning tasks was 0.0% accuracy, highlighting its nature as a base model not fine-tuned for such directives. Email classification showed a 33.3% accuracy (7th percentile), indicating a weakness in nuanced classification without specific fine-tuning. Ethics and general knowledge benchmarks yielded 50.0% and 36.4% accuracy respectively, both in the lower percentiles for accuracy but with excellent cost efficiency (98th percentile for both). The model's strengths lie in its foundational capabilities, speed, and cost-effectiveness, while its primary weakness is its inability to directly follow complex instructions or perform high-accuracy classification and reasoning without further fine-tuning, as expected from a base model.

Model Pricing

Current Pricing

Feature Price (per 1M tokens)
Prompt $0.2
Completion $0.8

Price History

Available Endpoints
Provider Endpoint Name Context Length Pricing (Input) Pricing (Output)
Chutes
Chutes | deepseek/deepseek-v3.1-base 163K $0.2 / 1M tokens $0.8 / 1M tokens
Benchmark Results
Benchmark Category Reasoning Free Executions Accuracy Cost Duration
Other Models by deepseek