DeepSeek: DeepSeek V3.1 Terminus

Text input Text output
Author's Description

DeepSeek-V3.1 Terminus is an update to [DeepSeek V3.1](/deepseek/deepseek-chat-v3.1) that maintains the model's original capabilities while addressing issues reported by users, including language consistency and agent capabilities, further optimizing the model's performance in coding and search agents. It is a large hybrid reasoning model (671B parameters, 37B active) that supports both thinking and non-thinking modes. It extends the DeepSeek-V3 base with a two-phase long-context training process, reaching up to 128K tokens, and uses FP8 microscaling for efficient inference. Users can control the reasoning behaviour with the `reasoning` `enabled` boolean. [Learn more in our docs](https://openrouter.ai/docs/use-cases/reasoning-tokens#enable-reasoning-with-default-config) The model improves tool use, code generation, and reasoning efficiency, achieving performance comparable to DeepSeek-R1 on difficult benchmarks while responding more quickly. It supports structured tool calling, code agents, and search agents, making it suitable for research, coding, and agentic workflows.

Key Specifications
Cost
$$$$
Context
131K
Parameters
671B (Rumoured)
Released
Sep 22, 2025
Speed
Ability
Reliability
Supported Parameters

This model supports the following parameters:

Top P Reasoning Tool Choice Temperature Stop Tools Max Tokens Presence Penalty Frequency Penalty Logprobs Include Reasoning Top Logprobs
Features

This model supports the following features:

Reasoning Tools
Performance Summary

DeepSeek-V3.1 Terminus, an update to DeepSeek-V3.1, demonstrates strong overall performance, particularly in reliability and specific accuracy benchmarks. It performs among the fastest models, ranking in the 67th percentile for speed across seven benchmarks, and offers competitive pricing, placing in the 46th percentile. Notably, the model exhibits exceptional reliability with a 100% success rate, indicating consistent and stable operation without technical failures. The model excels in critical areas, achieving perfect 100% accuracy in both Hallucinations (Baseline) and Ethics (Baseline) benchmarks, highlighting its ability to acknowledge uncertainty and adhere to ethical principles. It also shows robust performance in General Knowledge (99.5% accuracy) and Reasoning (90.0% accuracy), placing it in the 79th and 81st percentiles respectively. Its coding capabilities are strong, with 90.0% accuracy in the Coding (Baseline) benchmark. While its Email Classification accuracy (97.0%) is solid, it ranks in the 46th percentile, suggesting some room for improvement compared to top performers in this specific category. The model's hybrid reasoning architecture and optimized long-context training contribute to its efficiency in tool use, code generation, and agentic workflows.

Model Pricing

Current Pricing

Feature Price (per 1M tokens)
Prompt $0.21
Completion $0.79
Input Cache Read $0.168

Price History

Available Endpoints
Provider Endpoint Name Context Length Pricing (Input) Pricing (Output)
DeepSeek
DeepSeek | deepseek/deepseek-v3.1-terminus 131K $0.21 / 1M tokens $0.79 / 1M tokens
Novita
Novita | deepseek/deepseek-v3.1-terminus 131K $0.21 / 1M tokens $0.79 / 1M tokens
AtlasCloud
AtlasCloud | deepseek/deepseek-v3.1-terminus 131K $0.21 / 1M tokens $0.8 / 1M tokens
DeepInfra
DeepInfra | deepseek/deepseek-v3.1-terminus 163K $0.21 / 1M tokens $0.79 / 1M tokens
Chutes
Chutes | deepseek/deepseek-v3.1-terminus 163K $0.23 / 1M tokens $0.9 / 1M tokens
SiliconFlow
SiliconFlow | deepseek/deepseek-v3.1-terminus 163K $0.27 / 1M tokens $1 / 1M tokens
SambaNova
SambaNova | deepseek/deepseek-v3.1-terminus 131K $3 / 1M tokens $4.5 / 1M tokens
Novita
Novita | deepseek/deepseek-v3.1-terminus 131K $0.216 / 1M tokens $0.8 / 1M tokens
Benchmark Results
Benchmark Category Reasoning Strategy Free Executions Accuracy Cost Duration
Other Models by deepseek