Inference
EN
DeepSeek

DeepSeek V3.2 (Thinking)

DeepSeek V3.2 (Thinking) is a reasoning model from DeepSeek in the DeepSeek V3.2 family. 10 benchmarks count toward its index, in 4 categories.

  • Open weights
  • Reasoning
  • Input: text
  • Released 2025-12
RankedEnough results for an index.
IndexAll categories, out of 100.49.4±8.6Median 51.0
CoverageIndex weight that has results.65%
SpeedOutput tokens per second.—Median 67/s
First tokenSeconds to the first output token.—Median 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.55Median $0.60
Output / 1MUS dollars per 1M output tokens.$2.19Median $2.55
ContextMaximum tokens in one request.128KMedian 262K
EloLMArena rating and rank.1420±4 #116Median 1422

Run locally

685.36B parameters
16-bit
1440.3 GB
8-bit
765.6 GB
4-bit
437.3 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy needs more than 128 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.2 counted47.7151 / 218
CodingCode writing and repair.4 counted49.2143 / 209
ReasoningLogic problems and puzzles.None counted——
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.2 counted51.2125 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.None counted——
MathMath problems.2 counted51.082 / 168

Results

10 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
American Invitational Mathematics Examination 2025Math93.1%11 / 53
AIMEMath84.6% ±1.139 / 95
MMLU ProKnowledge84.9% ±0.459 / 135
LiveCodeBenchCoding80.7% ±1.165 / 142
Terminal-Bench 1.0Agentic40.0% ±5.522 / 47
GPQA Diamond · Vals AIKnowledge80.3% ±2.066 / 135
IOI v1Coding10.7% ±3.134 / 61
Terminal-Bench 2.0 · Vals AIAgentic36.0% ±5.141 / 66
SWE-benchCoding67.6% ±2.169 / 87
Vibe Code Bench v1.1Coding5.1% ±2.1109 / 123
MedQAHealthcare93.9% ±0.220 / 95
CaseLaw v2Legal55.4% ±0.534 / 53
TaxEval v2Finance68.2% ±0.9100 / 143
LegalBenchLegal76.1% ±0.5112 / 149
CorpFin v2Finance51.0% ±1.0105 / 131
MGSMMultilingual86.0% ±2.065 / 74
Finance Agent v1.1Finance2.3% ±0.849 / 51

Sources

BenchLMCC BY-NC 4.0Vals AINo licence stated

History

5 releases
Index50.5 → 49.4
23 Oct 202612 Nov 2026
Output / 1M$2.19 → $2.19
23 Oct 202612 Nov 2026

Questions

How much does DeepSeek V3.2 (Thinking) cost?
$0.55 per 1M input tokens, $2.19 per 1M output tokens.
How good is DeepSeek V3.2 (Thinking)?
Its index is 49.4 out of 100. That is place 153 of 283 ranked models.
What is DeepSeek V3.2 (Thinking) best at?
Knowledge: 51.2 out of 100, place 125 of 268.
What is the context window of DeepSeek V3.2 (Thinking)?
128K tokens.
Is DeepSeek V3.2 (Thinking) open source?
Yes. Its weights are published.
When was DeepSeek V3.2 (Thinking) released?
December 2025.

Badge

DeepSeek V3.2 (Thinking) on Inference360: place 153 of 283, index 49.4
HTML
<a href="https://inference360.ai/models/deepseek-v3-2-thinking"><img src="https://inference360.ai/badges/deepseek-v3-2-thinking.svg" alt="DeepSeek V3.2 (Thinking) on Inference360: place 153 of 283, index 49.4" height="20"></a>
Markdown
[![DeepSeek V3.2 (Thinking) on Inference360: place 153 of 283, index 49.4](https://inference360.ai/badges/deepseek-v3-2-thinking.svg)](https://inference360.ai/models/deepseek-v3-2-thinking)

Compare with

Gemini 3.1 Flash Lite Preview49.6MiniMax M2.149.1DeepSeek V3.244.2DeepSeek V3.137.9

Lower price

dots3-note Preview63.4 · FreeApodex 1.1 Mini55.7 · FreeNemotron 3 Ultra50.8 · Free

More from DeepSeek

All models
DeepSeek V3.244.2DeepSeekMath V2UnrankedDeepSeek V3.2 ExpUnrankedDeepSeek V3.1 TerminusUnrankedDeepSeek V3.1 (Reasoning)43.0