NOUMENOpen app
NOUMEN

DeepSeek · Open weights

DeepSeek V3.1 Terminus (Thinking)

Added Jan 1, 2024Open weights

Compare models

Choose up to 3 models to compare.0 of 3 chosen

Loading models…

Use in the app

Thinking-enabled DeepSeek-V3.1-Terminus with improved language consistency, upgraded Code/Search Agents, and stronger stability and reliability versus V3.1. FP8.

  • Tools
  • Structured output
Input
Text
Output
Text
Discovery tier
Open weights

Pricing

USD per 1M tokens
Input
$0.25
Output
$0.70

Reference prices as published by the model provider. API example

Specifications

Context window
128K
Max output
65.5K
Average output (7 days)
362 tokens
Longer than 37% of 465 models

Side by side

Compare models

Put DeepSeek V3.1 Terminus (Thinking) next to up to three other models. The comparison stays in the link, so you can share it.

Compare models

Choose up to 3 models to compare.0 of 3 chosen

Loading models…

TryGLM 4.7 Flash ThinkingGemma 4 31B IT TEEQwen3.6 35B A3B TEEAll 3

Public evaluations

Benchmarks

Independent evaluation results, with how each score ranks among the models in the Noumen catalog.

Intelligence Index

14.8

Better than 37% of 326 scored models

3.8Catalog models by score55.1

Indexes

  • Coding Index

    43.5

    Composite of the coding evaluations

    Better than 36% of 168 scored models

  • Math Index

    89.7

    Composite of the math evaluations

    Better than 87% of 118 scored models

Agentic work

  • AutomationBench-AA

    2.7%

    Workflow automation, penalized for guardrail violations

    Better than 17% of 173 scored models

  • AutomationBench-AA tasks completed

    0.2%

    Workflows finished end to end without a guardrail violation

    Better than 13% of 173 scored models

  • AA-Briefcasev1.1

    458 Elo

    Agentic knowledge work, rated head to head

    Better than 18% of 166 scored models

  • GDPval-AAv2.1

    709 Elo

    Economically valuable occupational tasks, rated head to head

    Better than 22% of 187 scored models

Document reasoning

  • GDP.pdf

    5.4%

    Professional reasoning over PDF documents, all-pass rate

    Better than 23% of 171 scored models

  • AA-LCRv1.1

    69.3%

    Reasoning across long documents

    Better than 48% of 309 scored models

Reasoning and knowledge

  • Humanity’s Last Exam

    16.4%

    Expert-level questions across academic fields

    Better than 43% of 323 scored models

  • GPQA Diamond

    79.2%

    Graduate-level science questions

    Better than 46% of 291 scored models

  • MMLU-Pro

    85.1%

    Multi-subject knowledge and reasoning

    Better than 87% of 132 scored models

  • IFBench

    57.0%

    Following precise output instructions

    Better than 57% of 215 scored models

Coding

  • Terminal-Benchv4.0

    0.0%

    Practical software engineering in a terminal

    Better than 0% of 171 scored models

  • SciCode

    38.0%

    Scientific computing in Python

    Better than 17% of 174 scored models

  • LiveCodeBench

    79.8%

    Recent competitive programming problems

    Better than 82% of 129 scored models

Math

  • AIME 2025

    89.7%

    Competition mathematics, 2025 problems

    Better than 87% of 118 scored models

Legacy evaluations

  • Terminal-Bench Hard

    30.3%

    Earlier terminal task suite

    Better than 56% of 212 scored models

  • τ²-Bench Telecom

    37.1%

    Tool use in a customer-support simulation

    Better than 31% of 214 scored models

  • AA-LCR (unversioned)

    69.3%

    Long-context reasoning, earlier grading

    Better than 48% of 309 scored models

Scores from Artificial Analysis, updated Oct 26, 2026. Artificial Analysis

Live routing data

Performance and reliability

Recent averages across requests routed for this model. Your own speed depends on prompt length, reasoning effort and load.

Throughput
22.3 tok/s
Faster than 14% of 250 models
Time to first token
8.0 s
Quicker than 16% of 250 models
Prompt cache hits
79.6%
Request success (24h)
99.5%
Reliable

Cache hits are the share of input tokens recently served from a prompt cache across all requests. They are not a forecast of your own hit rate.

Hourly request success · last 72 hoursLowest hour 91.6%
  • Healthy (98% or more)
  • Degraded (90–98%)
  • Down (under 90%)
  • No requests

Developers

Use it through the API

Noumen’s API is OpenAI-compatible: any Chat Completions client works with a base URL and a key.

Read the quickstart
curl https://www.noumen.xyz/api/v1/chat/completions \
  -H "Authorization: Bearer $NOUMEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-ai/DeepSeek-V3.1-Terminus:thinking",
    "messages": [{ "role": "user", "content": "Explain this stack trace." }]
  }'
import os
from openai import OpenAI

client = OpenAI(base_url="https://www.noumen.xyz/api/v1", api_key=os.environ["NOUMEN_API_KEY"])
reply = client.chat.completions.create(
    model="deepseek-ai/DeepSeek-V3.1-Terminus:thinking",
    messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)

DeepSeek

Other DeepSeek models, starting with the same family.