BenchLeader

DeepSeek V3.2

Reasoning effort
Quality#82 of 339
56.7Index
Top quarter
+6.5 on the median
Price#68 of 255
$0.315per 1M tokens
Cheaper than most
0.4× the median
Speed#184 of 216
40tokens/s
Bottom quarter
0.5× the median
Value#56 of 255
180.0points per $
Top quarter
3.3× the median

Better value: MiMo-V2.5 scores 56.3 for $0.138/M, 56% cheaper.

2 cheaper ranked models score higher: GLM 5.3 Flash (57.8, $0.238), DeepSeek V4 Flash (57.2, $0.262).

As of 15 Nov 2026, DeepSeek V3.2 by DeepSeek scores 56.7 on the BenchLeader Index (608 ranked configurations), with its best results in agents & tools (66) and weakest in maths (49). API pricing: $0.280 per 1M input tokens and $0.420 per 1M output tokens ($0.315 blended at 3:1), list price on 15 Nov 2026. It streams about 40 tokens/s (Artificial Analysis). Released 1 Dec 2025.

Key facts (as of 15 Nov 2026)

  • DeepSeek V3.2 API pricing: $0.280 per 1M input tokens and $0.420 per 1M output tokens ($0.315 blended at 3:1) (list price, 15 Nov 2026).
  • DeepSeek V3.2 streams about 40 tokens/s (Artificial Analysis, 15 Nov 2026).
  • DeepSeek V3.2 was released on 1 Dec 2025 (DeepSeek).

In detail. DeepSeek V3.2 is a DeepSeek open-weights reasoning model, released 1 Dec 2025. Its best configuration (thinking reasoning effort) ranks #170 of 608 on the BenchLeader Index at 56.7 ±4.4, in the upper half. The ± is the point: 266 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in agents & tools (66) and lowest in maths (49). At $0.315 per million tokens blended it is cheaper than most ranked models. Output speed of 40 tokens per second puts it in the slowest quarter, with a first answer in 29.4 s. It has been measured at 4 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 11 Nov 2026.

Blended price
$0.315/M
$0.280 in · $0.420 out
via Artificial Analysis
Output speed
40 tok/s
measured by Artificial Analysis
First answer
29 s
first token 1.13 s
Context
131k
via LiteLLM price table
Full answer
84 s
last measured 13 Nov 2026
Cost per task
–
Tokens per task
–
Released
1 Dec 2025
How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.
  1. Overall index57
  2. Reasoning60
  3. Coding52
  4. Agents & tools66
  5. Maths49
  6. Knowledge53
  7. Instruction following61
  8. Human preference60
  9. Long context61

Embed this rank

More badge styles
[![DeepSeek V3.2 on BenchLeader: #81 overall](https://www.benchleader.com/badge/deepseek-v3-2.svg?style=compact)](https://www.benchleader.com/models/deepseek-v3-2)

Paste it at the top of README.md, beside your other badges. The badge redraws with every data update and links back to this page.

Get an email when DeepSeek ships a new model

And once when it gets ranked, plus the weekly review on Mondays.

Choose providers and emails

Versions

DeepSeek has shipped 2 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.

ModelReleasedIndexRank
DeepSeek V3.2thinkingthis page1 Dec 202556.7#170
DeepSeek V3.1thinking21 Aug 202552.4#278

Reasoning-effort configurations

The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.

EffortIndexRankSpeedFirst answerChat reply costCategories
no reasoning52.1#28554 tok/s2.87 s$0.0002Agents & tools 62 · Coding 63 · Instruction following 51 · Knowledge 44 · Long context 47 · Maths 51
high54.2#24028 tok/s1.13 s$0.0002Coding 61 · Knowledge 49 · Maths 60
thinkingbest56.7#17040 tok/s29 s$0.0002Agents & tools 66 · Coding 52 · Human preference 60 · Instruction following 61 · Knowledge 53 · Long context 61 · Maths 49 · Reasoning 60
not stated50.2#34128 tok/s1.13 s$0.0002Agents & tools 50 · Coding 43 · Human preference 60 · Knowledge 46 · Maths 54 · Reasoning 49

How its index has moved

8 Sep 2026 to 15 Nov 2026
465462
  • thinking
  • high
  • not stated

The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.

Benchmark results

See all LLM benchmarks. One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.

Metadata

BenchmarkthinkingSourceTrend
AA Openness Indexnot in index44.4#97Artificial Analysis

Reasoning

Benchmarkno reasoninghighthinkingnot statedSourceTrend
GPQA Diamondeffort not stated–––83.4%#113Epoch AI Benchmarking Hub
Arena Hard Prompts––1446#1191447#117Arena (LMArena)
GPQA Diamond (AA)not in index75.0%#255–84.0%#153–Artificial Analysis
Humanity's Last Exam (AA)not in index11.2%#316–24.6%#204–Artificial Analysis
GPQA Diamond (Vals)not in index76.3%#7680.3%#67––Vals AI
Kagi LLM Benchmarkeffort not stated–––52.2%#77Kagi LLM Benchmark
ARC-AGI-1effort not stated–––57.0%#176ARC Prize
ARC-AGI-2effort not stated–––4.0%#193ARC Prize
CritPtnot in index0.9%#256–2.9%#201–Artificial Analysis
Arena Expertnot in index––1454#1111448#121Arena (LMArena)
CL-benchnot in index––13.2%#2012.4%#21Epoch AI Benchmarking Hub
CL-bench Lifenot in index––9.5%#147.4%#15Epoch AI Benchmarking Hub
LMCAnot in index––29.1%#14828.8%#150Epoch AI Benchmarking Hub
DTBenchnot in index––87.7%#10385.6%#113Epoch AI Benchmarking Hub

Coding

Benchmarkno reasoninghighthinkingnot statedSourceTrend
SciCode––38.9%#190–SciCode
WeirdML––39.5%#106–WeirdML
Arena Coding––1475#1161469#125Arena (LMArena)
Arena WebDev––1363#1151326#128Arena (LMArena)
LiveCodeBench––80.7%#6569.9%#90Vals AI
SWE-bench (Vals)not in index–67.6%#68––Vals AI
SWE-Bench Proeffort not stated–––15.6%#20Scale AI SEAL
Aider Polyglot70.2%#16–74.2%#11–Aider polyglot leaderboard
SWE-bench Verified (bash only)–70.0%#13––SWE-bench
SWE-bench Multilingual (bash only)not in indexeffort not stated–––59.0%#12SWE-bench
SWE-bench Verified (any scaffold)not in index–70.0%#15––SWE-bench
Vibe Code Bench v1.1not in index–5.1%#110––Vals AI
Terminal-Bench 2.1 (AA)not in index––46.8%#128–Artificial Analysis
Design Arena Websitenot in indexeffort not stated–––1171#107Design Arena
LiveCodeBench (AA)not in index59.3%#111–86.2%#11–Artificial Analysis

Agents & tools

Benchmarkno reasoninghighthinkingnot statedSourceTrend
Terminal-Bencheffort not stated–––39.5%#56Terminal-Bench
APEX-Agentseffort not stated–––21.3%#64Mercor
Terminal-Bench Hard32.6%#98–35.6%#74–Artificial Analysis
τ²-Bench Telecom (AA)not in index79.0%#130–90.6%#62–Artificial Analysis
BFCL Overall––56.7%#1254.1%#17Berkeley Function Calling Leaderboard
GDPval-AA v2.1not in index––699#223–Artificial Analysis
APEX-Agents (AA)not in index––14.5%#24–Artificial Analysis
Poker Agentnot in index––1090.3%#4–Vals AI
Terminal-Bench 2.0 (Vals)not in index34.8%#4136.0%#40––Vals AI
TheAgentCompanynot in indexeffort not stated–––42.9%#1TheAgentCompany
Vending-Bench 2not in indexeffort not stated–––1034#58Andon Labs

Maths

Benchmarkno reasoninghighthinkingnot statedSourceTrend
OTIS Mock AIMEeffort not stated–––87.8%#93Epoch AI Benchmarking Hub
ProofBencheffort not stated–––8.0%#74Vals AI
AIME (Vals)64.8%#5584.6%#39––Vals AI
MGSM––86.0%#6490.9%#32Vals AI
AIME 2026––94.2%#20–MathArena
HMMT February 2026––84.1%#25–MathArena
MathArena Apexpossible contamination––2.1%#29–MathArena
Arena Maths––1429#1091426#114Arena (LMArena)
AIME 2025 (AA)not in index59.0%#114–92.0%#17–Artificial Analysis

Knowledge

Benchmarkno reasoninghighthinkingnot statedSourceTrend
MMLU-Pro––84.9%#5983.1%#72Vals AI
LegalBench––76.1%#11066.0%#133Vals AI
CorpFin47.9%#10751.0%#102––Vals AI
TaxEval–68.2%#98––Vals AI
MedQA89.5%#5193.9%#20––Vals AI
CaseLaw v2not in index–55.4%#34––Vals AI
Global-MMLU-Lite (AA)not in index82.7%#54–86.5%#30–Artificial Analysis

Instruction following

Benchmarkno reasoningthinkingnot statedSourceTrend
IFBench49.0%#18360.7%#126–Artificial Analysis
Arena Instruction Followingnot in index–1420#1121421#111Arena (LMArena)

Human preference

Benchmarkthinkingnot statedSourceTrend
Arena Text1425#1211425#122Arena (LMArena)
EQ-Bench Creative Writing v3not in indexeffort not stated–1515#57EQ-Bench
Arena Creative Writingnot in index1391#1241408#95Arena (LMArena)
Arena Multi-turnnot in index1428#1221428#121Arena (LMArena)
Arena Longer Queriesnot in index1442#1111441#112Arena (LMArena)

Long context

Benchmarkno reasoningthinkingSourceTrend
AA-LCR v1.145.7%#35673.3%#195Artificial Analysis

Composite

Benchmarkno reasoningthinkingnot statedSourceTrend
Epoch Capabilities Indexnot in indexeffort not stated––146.3#86Epoch AI Benchmarking Hub
AA Intelligence Index v4.3.2not in index–21.5#221–Artificial Analysis
AA Progress Index v1not in index226.2#329264.9#223–Artificial Analysis

API pricing: where to run it

Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.

ProviderSpeedFirst tokenInput $/MOutput $/MContextQuantisation
GMICloud48 tok/s1.39 s$0.209$0.310164kfp8
Baidu Qianfan32 tok/s1.31 s$0.280$0.420131kfp8
DeepInfra28 tok/s0.68 s$0.260$0.380164kfp4
DigitalOcean25 tok/s2.08 s$0.300$0.960164k–
AtlasCloud24 tok/s2.86 s$0.260$0.380164kfp8
SiliconFlow18 tok/s2.07 s$0.259$0.800164kfp8

Usage

How much DeepSeek V3.2 is used in practice, as the gateways that publish it report. Usage is not part of the Index. Most used models and labs.

OpenRouter, past week
#59
0.14% of tokens

Source: Usage: OpenRouter rankings · data as of 15 Nov 2026.

What a task costs

Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache at $0.130 per 1M. Reasoning tokens are not modelled.

WorkloadTokens in / outCostWith cachingTime
Chat reply400 / 300$0.0002$0.000236.9 s
Summarize a 30-page report12,000 / 600$0.0036$0.002344.3 s
Code edit6,000 / 1,500$0.0023$0.00161.1 min
Agentic coding session60,000 / 4,000$0.018$0.0122.1 min
Structured extraction2,000 / 200$0.0006$0.000434.4 s

More from DeepSeek

New this month

See also

Embed or cite

Show where DeepSeek V3.2 ranks on a model card or README. The badge redraws itself with every daily update, carries the data date and links back to this page.

[![DeepSeek V3.2 on BenchLeader: #81 overall](https://www.benchleader.com/badge/deepseek-v3-2.svg)](https://www.benchleader.com/models/deepseek-v3-2)

Cite this

The ranking changes daily, so a reference carries the data date. APA style:

BenchLeader. (2026, November 15). DeepSeek V3.2 on the BenchLeader Index (method v1.4) [Data set]. Retrieved November 15, 2026, from https://www.benchleader.com/models/deepseek-v3-2

BibTeX:

@misc{benchleader_deepseek_v3_2_20261115,
  author       = {{BenchLeader}},
  title        = {{DeepSeek V3.2 on the BenchLeader Index (method v1.4)}},
  year         = {2026},
  month        = nov,
  howpublished = {\url{https://www.benchleader.com/models/deepseek-v3-2}},
  note         = {Data as of 2026-11-15}
}

Every badge style, the embeddable top-10 widget and how to cite the Index

Questions people ask

What are DeepSeek V3.2's benchmark scores?
As of 15 Nov 2026, DeepSeek V3.2 (thinking effort) scores 56.7 ±4.4 on the BenchLeader Index, strongest in agents & tools (66) and weakest in maths (49). That rests on independent benchmark results only, and every score on this page links to its publisher.
How does DeepSeek V3.2 rank?
DeepSeek V3.2 (thinking effort) is #170 of 608 ranked configurations on 15 Nov 2026. 266 other configurations score within its margin, so the exact position matters less than the band. The leader is Claude Fable 5.2 (max effort) at 72.7.
How much does DeepSeek V3.2 cost per 1M tokens?
As of 15 Nov 2026, DeepSeek V3.2 costs $0.280 per 1M input tokens and $0.420 per 1M output tokens, $0.315 blended (three input tokens per output token).
When was DeepSeek V3.2 released?
DeepSeek released DeepSeek V3.2 on 1 Dec 2025. BenchLeader has results for it at 4 reasoning-effort settings.

Data as of 15 Nov 2026. Compare these configurations.

Cite as: BenchLeader, “DeepSeek V3.2: benchmarks, pricing, speed and rank”, data as of 15 Nov 2026, https://www.benchleader.com/models/deepseek-v3-2.