NOUMENOpen app
NOUMEN

DeepSeek · Open weights

DeepSeek V3.2 Exp Thinking

Added Jan 1, 2024Open weights

Compare models

Choose up to 3 models to compare.1 of 3 chosen

Loading models…

Use in the app

DeepSeek V3.2 Exp Thinking is an experimental V3-series reasoning model with improved long-context efficiency compared with earlier V3 releases. FP8.

  • Reasoning
Input
Text
Output
Text
Discovery tier
Open weights

Pricing

USD per 1M tokens
Input
$0.28
Output
$0.42

Reference prices as published by the model provider. API example

Specifications

Context window
163.8K
Max output
65.5K
Average output (7 days)
917 tokens
Longer than 68% of 468 models

Side by side

DeepSeek V3.2 Exp Thinking compared

Specifications, prices, speed and benchmark scores next to each other. Benchmarks below show every compared model too.

Compare models

Choose up to 3 models to compare.1 of 3 chosen

Loading models…

DeepSeek V3.2 Exp ThinkingDeepSeekThis modelGemma 4 26B A4BGoogleRemove
ProviderDeepSeekGoogle
Discovery tierOpen weightsOpen weights
AddedJan 1, 2024Apr 2, 2026
Limits
Context window163.84K262.14KBest
Max output65.54K131.07KBest
Average output917 tokens131 tokens
Pricing, USD per 1M tokens
Input$0.28$0.09Best
Output$0.42$0.30Best
Cache read—$0.05
Performance
Throughput12.8 tok/s65.0 tok/sBest
Time to first token5.0 s2.3 sBest
Request success (24h)100.0%100.0%
Benchmark indexes
Intelligence Index16.616.7Best
Coding Index—39.3
Math Index87.7—
Capabilities
ReasoningYesYes
ToolsNoYes
VisionNoYes
Structured outputNoYes
Video inputNoYes
Reasoning effort——

Public evaluations

Benchmarks

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

  • DeepSeek V3.2 Exp Thinking
  • Gemma 4 26B A4B

Intelligence Index

  • DeepSeek V3.2 Exp Thinking16.6
  • Gemma 4 26B A4B16.7

DeepSeek V3.2 Exp Thinking: Better than 43% of 326 scored models

3.8Catalog models by score55.1

Indexes

  • Coding Index

    Composite of the coding evaluations

    • DeepSeek V3.2 Exp Thinking: —
    • Gemma 4 26B A4B: 39.3

  • Math Index

    Composite of the math evaluations

    • DeepSeek V3.2 Exp Thinking: 87.7
    • Gemma 4 26B A4B: —

Agentic work

  • AutomationBench-AA

    Workflow automation, penalized for guardrail violations

    • DeepSeek V3.2 Exp Thinking: —
    • Gemma 4 26B A4B: 1.9%

  • AutomationBench-AA tasks completed

    Workflows finished end to end without a guardrail violation

    • DeepSeek V3.2 Exp Thinking: —
    • Gemma 4 26B A4B: 0.2%

  • AA-Briefcasev1.1

    Agentic knowledge work, rated head to head

    • DeepSeek V3.2 Exp Thinking: —
    • Gemma 4 26B A4B: 315 Elo

  • GDPval-AAv2.1

    Economically valuable occupational tasks, rated head to head

    • DeepSeek V3.2 Exp Thinking: —
    • Gemma 4 26B A4B: 565 Elo

Document reasoning

  • AA-LCRv1.1

    Reasoning across long documents

    • DeepSeek V3.2 Exp Thinking: 72.3%Top
    • Gemma 4 26B A4B: 65.7%

Reasoning and knowledge

  • Humanity’s Last Exam

    Expert-level questions across academic fields

    • DeepSeek V3.2 Exp Thinking: 14.9%
    • Gemma 4 26B A4B: 19.3%Top

  • GPQA Diamond

    Graduate-level science questions

    • DeepSeek V3.2 Exp Thinking: 79.7%Top
    • Gemma 4 26B A4B: 79.2%

  • MMLU-Pro

    Multi-subject knowledge and reasoning

    • DeepSeek V3.2 Exp Thinking: 85.0%
    • Gemma 4 26B A4B: —

  • IFBench

    Following precise output instructions

    • DeepSeek V3.2 Exp Thinking: 54.1%
    • Gemma 4 26B A4B: 72.4%Top

Coding

  • SciCode

    Scientific computing in Python

    • DeepSeek V3.2 Exp Thinking: —
    • Gemma 4 26B A4B: 40.0%

  • LiveCodeBench

    Recent competitive programming problems

    • DeepSeek V3.2 Exp Thinking: 78.9%
    • Gemma 4 26B A4B: —

Math

  • AIME 2025

    Competition mathematics, 2025 problems

    • DeepSeek V3.2 Exp Thinking: 87.7%
    • Gemma 4 26B A4B: —

Legacy evaluations

  • Terminal-Bench Hard

    Earlier terminal task suite

    • DeepSeek V3.2 Exp Thinking: 31.1%Top
    • Gemma 4 26B A4B: 13.6%

  • τ²-Bench Telecom

    Tool use in a customer-support simulation

    • DeepSeek V3.2 Exp Thinking: 33.9%
    • Gemma 4 26B A4B: 43.6%Top

  • AA-LCR (unversioned)

    Long-context reasoning, earlier grading

    • DeepSeek V3.2 Exp Thinking: 72.3%Top
    • Gemma 4 26B A4B: 65.7%

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
12.8 tok/s
Faster than 2% of 259 models
Time to first token
5.0 s
Quicker than 33% of 259 models
Prompt cache hits
79.0%
Request success (24h)
100.0%
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 98.0%
  • 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-v3.2-exp-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-v3.2-exp-thinking",
    messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)

DeepSeek

Other DeepSeek models, starting with the same family.