Models / DeepSeek-V3.2-Exp

DeepSeek

DeepSeek-V3.2-Exp

DeepSeek-V3.2-Exp is strongest at sticking to the facts; worth a look for simple, high-volume work where cost matters most. The cheapest model of similar intelligence is Qwen3.5-9B, at $0.11 per million tokens against $0.30.

Superseded. DeepSeek-V3.2 is DeepSeek's newer model in this line, at $0.32 blended: see DeepSeek-V3.2 or compare the two.

Results 16 Nov 2026 · 53 results · 6 sources

Test DeepSeek-V3.2-Exp on your prompt → It runs beside the models you could use instead, with every answer checked and costed.

Price per million tokens
$0.27 in · $0.41 out
Typical host on OpenRouter, 10 Nov 2026 · compare prices
Speed
Not measured
Intelligence Index
16.6
Artificial Analysis, at thinking reasoning; 13.9–16.6 across 2 settings
Spring Prompt overall
Not ranked yet
needs our benchmarks and two groups of results
Developer
DeepSeek, based in China
Weights published
Released
29 Sep 2025
163,840 tokens of context

Where it stands among the models you could choose

Available models grouped by price, intelligence, speed and developer location, plus models with published weights. Retired models are excluded.

AmongIntelligence IndexSpring Prompt overallPrice, blendedSpeed
Models available today 107th of 185Claude Opus 5.5 – $0.30 blended, 27th percentilecheapest: Mistral Nemo –
Models priced $0.25–1 per million tokens 29th of 61MiMo-V2.6-Pro – sets this group –
Models with similar intelligence sets this group – $0.30 blended, 25th percentilecheapest: Qwen3.5-9B –
Models from developers based in China 45th of 75MiMo-V2.6-Pro – $0.30 blended, 24th percentilecheapest: Qwen3.5-9B –
Open-weights models 46th of 104MiMo-V2.6-Pro – $0.30 blended, 38th percentilecheapest: Mistral Nemo –

The name under each rank is the leader of that group; green is the top quarter of the group and red the bottom quarter, by rank. Price is shown as a percentile: the share of the group that costs less, so lower is cheaper. "Sets this group" marks the measure the group is defined by. Spring Prompt overall is our score out of 100 across every source (how it works). Price is the developer's list price (or the typical OpenRouter host where we haven't read one) for three input tokens to one output; intelligence and speed are from Artificial Analysis (data sourced from Artificial Analysis); "similar intelligence" means within 4 points on its Intelligence Index. Developer locations are where each company is based, not where a model is served. Groups under 3 models are not ranked.

How good is it, and for what?

Each use case ranks the available models its benchmarks measured, then those priced $0.25–1 per million tokens. The bar shows its percentile in that field, best to the right. Open a row for the results behind it.

Product listingsTurning a sparse product feed and photos into listings that can go live Not measured
  • CatalogBench: has not measured this model.
Decks from an analysisTurning a finished analysis into a deck you could present as it is Not measured
  • DeckBench: has not measured this model.
User surveysPlanning a user survey and reading its results without being misled Not measured
  • SurveyBench: has not measured this model.
Marketing planningPlanning a year of ad spend without overspending Not measured
  • ROASBench: has not measured this model.
Agents and tool useMulti-step tasks with tools: support desks, coding agents, function calls 76th of 159 available · 13th of 47 at its price
BenchmarkDeepSeek-V3.2-ExpRank among availableBest available
Berkeley Function Calling Leaderboard (BFCL) V4BFCL: overall accuracy · reported 56.7% 7th of 36 Claude Opus 4.5 77.5%
Vending-Bench 2Vending-Bench 2: money after a year · reported $1,034 41st of 53 GPT-6 Astra $15,515
  • tau2-bench: has not measured this model.
  • OpenHands Index: has not measured this model.
  • Microsoft STATE-Bench: has not measured this model.
  • Artificial Analysis: has not measured this model.
Professional workReal tasks from banking, consulting and law, business documents and freelance projects Not measured
  • APEX-Agents: has not measured this model.
  • GDP.pdf: has not measured this model.
  • Remote Labor Index: has not measured this model.
Reasoning and knowledgeHard questions across science, maths and general knowledge 111th of 187 available · 29th of 61 at its price
BenchmarkDeepSeek-V3.2-ExpRank among availableBest available
Artificial AnalysisArtificial Analysis: Artificial Analysis Intelligence Index · reported 16.6 107th of 185 Claude Opus 5.5 57.6
Artificial AnalysisArtificial Analysis: Humanity's Last Exam · reported 14.9% 114th of 186 Claude Opus 5.5 61.4%
Artificial AnalysisArtificial Analysis: GPQA Diamond · reported 79.7% 95th of 168 GPT-6 Astra 96.3%
WritingWhat people prefer in blind comparisons, and judged writing quality 71st of 169 available · 12th of 58 at its price
BenchmarkDeepSeek-V3.2-ExpRank among availableBest available
Arena (formerly LMArena)Arena Text: overall · reported 1,425 75th of 153 Gemini 4 Argon 1,525
UGI LeaderboardUGI: writing score · reported 54.0 48th of 124 Gemini 3.8 Flash 78.6
Sticking to the factsSummarising without inventing things, and factual answers 40th of 139 available · 6th of 40 at its price
BenchmarkDeepSeek-V3.2-ExpRank among availableBest available
Vectara Hallucination LeaderboardVectara: hallucination rate · reported 5.3% 9th of 71 GPT-5.4 nano 3.1%
Arena (formerly LMArena)Arena Text factuality: overall · reported 1,446 55th of 109 Gemini 4 Argon 1,511
  • SimpleQA Verified (Epoch AI): has not measured this model.
Speed under pressureGood decisions against a real clock (fast chess) Not measured
  • BulletBench: has not measured this model.

Nearest alternatives

Against DeepSeek-V3.2-Exp at $0.30 blended, Intelligence Index 16.6. "Similar intelligence" means within 4 points or better.

Where to run it

1 host on OpenRouter at the standard service tier, with a median of $0.30 blended per million tokens; quantisation where the host states it: FP8. Cheapest first.

HostQuantisationInputOutputBlended
SiliconFlowFP8$0.27$0.41$0.30

Prices per million tokens from OpenRouter's public endpoint list, checked 10 Nov 2026.

Against the alternatives

The models you would most likely weigh it against: the leaders of the groups above.

Scroll sideways to see every alternative.

DeepSeek-V3.2-ExpMiMo-V2.6-Pro
most intelligent at $0.25–1 per million tokens
Qwen3.5-9B
cheapest with similar intelligence
Claude Opus 5.5
most intelligent available today
Price per million tokens $0.30 $0.54$0.11$8.00
Tokens a second – 4311096
Intelligence Index 16.6 46.313.357.6
Spring Prompt overall – ––82
Rank among available models, by use case
Product listings – ––13th
Decks from an analysis – ––7th
User surveys – ––4th
Marketing planning – ––9th
Agents and tool use 76th of 159 23rd111th3rd
Professional work – ––3rd
Reasoning and knowledge 111th of 187 13th116th1st
Writing 71st of 169 23rd112th5th
Sticking to the facts 40th of 139 10th–3rd
Speed under pressure – –––

Shaded figures are better than DeepSeek-V3.2-Exp's. A dash means no result.

Where it has been measured

29th most intelligent of the 61 models priced $0.25–1 per million tokens. Among models available today, mid-field on every use case measured so far.

Artificial Analysis lists it but has not published a speed for it yet

Every result

53 published results from 6 sources, each in the source's own units, with the configuration that produced it.

Arena (formerly LMArena) · reported by Arena (formerly LMArena) · 14 results
MeasureValueRankConfigurationDated
Arena Text factuality: overallArena rating, higher is better 1,436
1,430–1,442
100 of 182 DeepSeek-V3.2-Exp 2 Oct 2026
Arena Text factuality: overallArena rating, higher is better 1,446
1,435–1,457
77 of 182 DeepSeek-V3.2-Exp (thinking reasoning) 2 Oct 2026
Arena Text: business, management and financeArena rating, higher is better 1,422
1,410–1,435
120 of 406 DeepSeek-V3.2-Exp 2 Oct 2026
Arena Text: business, management and financeArena rating, higher is better 1,406
1,392–1,420
149 of 406 DeepSeek-V3.2-Exp (thinking reasoning) 2 Oct 2026
Arena Text: creative writingArena rating, higher is better 1,410
1,395–1,424
92 of 411 DeepSeek-V3.2-Exp 2 Oct 2026
Arena Text: creative writingArena rating, higher is better 1,391
1,374–1,409
123 of 411 DeepSeek-V3.2-Exp (thinking reasoning) 2 Oct 2026
Arena Text: expert promptsArena rating, higher is better 1,425
1,403–1,447
150 of 363 DeepSeek-V3.2-Exp 2 Oct 2026
Arena Text: expert promptsArena rating, higher is better 1,449
1,420–1,477
115 of 363 DeepSeek-V3.2-Exp (thinking reasoning) 2 Oct 2026
Arena Text: instruction followingArena rating, higher is better 1,414
1,404–1,425
121 of 413 DeepSeek-V3.2-Exp 2 Oct 2026
Arena Text: instruction followingArena rating, higher is better 1,416
1,404–1,427
115 of 413 DeepSeek-V3.2-Exp (thinking reasoning) 2 Oct 2026
Arena Text: overallArena rating, higher is better 1,422
1,416–1,428
127 of 413 DeepSeek-V3.2-Exp 2 Oct 2026
Arena Text: overallArena rating, higher is better 1,425
1,418–1,432
121 of 413 DeepSeek-V3.2-Exp (thinking reasoning) 2 Oct 2026
Arena Text: writing, literature and languageArena rating, higher is better 1,411
1,399–1,422
105 of 412 DeepSeek-V3.2-Exp 2 Oct 2026
Arena Text: writing, literature and languageArena rating, higher is better 1,398
1,385–1,411
132 of 412 DeepSeek-V3.2-Exp (thinking reasoning) 2 Oct 2026

Not shown: Not an error rate: claims that cannot be checked on the web are skipped, and preference still carries most of the weight.

Artificial Analysis · reported by Artificial Analysis · 14 results
MeasureValueRankConfigurationDated
Artificial Analysis: Artificial Analysis Intelligence Indexindex score, higher is better 13.9 267 of 413 DeepSeek-V3.2-Exp (no reasoning) 20 Oct 2026
Artificial Analysis: Artificial Analysis Intelligence Indexindex score, higher is better 16.6 240 of 413 DeepSeek-V3.2-Exp (thinking reasoning) 20 Oct 2026
Artificial Analysis: GPQA Diamond% of questions, higher is better 73.8% 226 of 360 DeepSeek-V3.2-Exp (no reasoning) 20 Oct 2026
Artificial Analysis: GPQA Diamond% of questions, higher is better 79.7% 177 of 360 DeepSeek-V3.2-Exp (thinking reasoning) 20 Oct 2026
Artificial Analysis: Humanity's Last Exam% of questions, higher is better 9.0% 289 of 419 DeepSeek-V3.2-Exp (no reasoning) 20 Oct 2026
Artificial Analysis: Humanity's Last Exam% of questions, higher is better 14.9% 233 of 419 DeepSeek-V3.2-Exp (thinking reasoning) 20 Oct 2026
Artificial Analysis: IFBench% of instructions, higher is better 43.1% 183 of 288 DeepSeek-V3.2-Exp (no reasoning) 20 Oct 2026
Artificial Analysis: IFBench% of instructions, higher is better 54.2% 124 of 288 DeepSeek-V3.2-Exp (thinking reasoning) 20 Oct 2026
Artificial Analysis: Long-context reasoning (AA-LCR)% of questions, higher is better 44.7% 300 of 407 DeepSeek-V3.2-Exp (no reasoning) 20 Oct 2026
Artificial Analysis: Long-context reasoning (AA-LCR)% of questions, higher is better 72.3% 178 of 407 DeepSeek-V3.2-Exp (thinking reasoning) 20 Oct 2026
Artificial Analysis: Terminal-Bench Hard% of tasks, higher is better 25.0% 128 of 282 DeepSeek-V3.2-Exp (no reasoning) 20 Oct 2026
Artificial Analysis: Terminal-Bench Hard% of tasks, higher is better 31.1% 103 of 282 DeepSeek-V3.2-Exp (thinking reasoning) 20 Oct 2026
Artificial Analysis: τ²-bench telecom% of tasks, higher is better 33.9% 197 of 286 DeepSeek-V3.2-Exp (no reasoning) 20 Oct 2026
Artificial Analysis: τ²-bench telecom% of tasks, higher is better 33.9% 197 of 286 DeepSeek-V3.2-Exp (thinking reasoning) 20 Oct 2026

Not shown: Not business work, and a blend: read the parts for any one task.

Berkeley Function Calling Leaderboard (BFCL) V4 · reported by Berkeley Function Calling Leaderboard (BFCL) V4 · 16 results
MeasureValueRankConfigurationDated
BFCL: irrelevance detection% correct, higher is better 93.2% 8 of 109 DeepSeek-V3.2-Exp (bfcl fc) 16 Dec 2025
BFCL: irrelevance detection% correct, higher is better 67.0% 86 of 109 DeepSeek-V3.2-Exp (thinking reasoning, bfcl prompt) 16 Dec 2025
BFCL: memory% correct, higher is better 54.2% 8 of 109 DeepSeek-V3.2-Exp (bfcl fc) 16 Dec 2025
BFCL: memory% correct, higher is better 44.1% 16 of 109 DeepSeek-V3.2-Exp (thinking reasoning, bfcl prompt) 16 Dec 2025
BFCL: multi-turn tasks% correct, higher is better 37.4% 31 of 109 DeepSeek-V3.2-Exp (bfcl fc) 16 Dec 2025
BFCL: multi-turn tasks% correct, higher is better 44.9% 22 of 109 DeepSeek-V3.2-Exp (thinking reasoning, bfcl prompt) 16 Dec 2025
BFCL: overall accuracy% correct, higher is better 54.1% 19 of 109 DeepSeek-V3.2-Exp (bfcl fc) 16 Dec 2025
BFCL: overall accuracy% correct, higher is better 56.7% 14 of 109 DeepSeek-V3.2-Exp (thinking reasoning, bfcl prompt) 16 Dec 2025
BFCL: relevance detection% correct, higher is better 37.5% 103 of 109 DeepSeek-V3.2-Exp (bfcl fc) 16 Dec 2025
BFCL: relevance detection% correct, higher is better 93.8% 8 of 109 DeepSeek-V3.2-Exp (thinking reasoning, bfcl prompt) 16 Dec 2025
BFCL: single-turn calls (curated)% correct, higher is better 34.9% 105 of 109 DeepSeek-V3.2-Exp (bfcl fc) 16 Dec 2025
BFCL: single-turn calls (curated)% correct, higher is better 85.5% 39 of 109 DeepSeek-V3.2-Exp (thinking reasoning, bfcl prompt) 16 Dec 2025
BFCL: single-turn calls (user-contributed)% correct, higher is better 53.7% 97 of 109 DeepSeek-V3.2-Exp (bfcl fc) 16 Dec 2025
BFCL: single-turn calls (user-contributed)% correct, higher is better 76.0% 35 of 109 DeepSeek-V3.2-Exp (thinking reasoning, bfcl prompt) 16 Dec 2025
BFCL: web search% correct, higher is better 69.5% 16 of 109 DeepSeek-V3.2-Exp (bfcl fc) 16 Dec 2025
BFCL: web search% correct, higher is better 58.0% 22 of 109 DeepSeek-V3.2-Exp (thinking reasoning, bfcl prompt) 16 Dec 2025

Not shown: Not general refusal or safety behaviour.

UGI Leaderboard · reported by UGI Leaderboard · 6 results
MeasureValueRankConfigurationDated
UGI: requested-length error% off the requested word count, lower is better 27.0% 247 of 379 DeepSeek-V3.2-Exp (no reasoning) 2 Oct 2025
UGI: requested-length error% off the requested word count, lower is better 25.0% 236 of 379 DeepSeek-V3.2-Exp (thinking reasoning) 2 Oct 2025
UGI: style adherencescore from 0 to 1, higher is better 0.35 205 of 379 DeepSeek-V3.2-Exp (no reasoning) 2 Oct 2025
UGI: style adherencescore from 0 to 1, higher is better 0.35 205 of 379 DeepSeek-V3.2-Exp (thinking reasoning) 2 Oct 2025
UGI: writing scorescore out of 100, higher is better 51.0 163 of 379 DeepSeek-V3.2-Exp (no reasoning) 2 Oct 2025
UGI: writing scorescore out of 100, higher is better 54.0 139 of 379 DeepSeek-V3.2-Exp (thinking reasoning) 2 Oct 2025

Not shown: Not other format limits such as character counts or bullet counts.

Vectara Hallucination Leaderboard · reported by Vectara Hallucination Leaderboard · 2 results
MeasureValueRankConfigurationDated
Vectara: answer rate% of documents, higher is better 96.6% 89 of 108 DeepSeek-V3.2-Exp 22 Sep 2026
Vectara: hallucination rate% of summaries, lower is better 5.3% 15 of 108 DeepSeek-V3.2-Exp 22 Sep 2026

Not shown: Not a quality score: a low rate usually means content filters were triggered, and hallucination rates are measured on answered documents only.

Vending-Bench 2 · reported by Vending-Bench 2 · 1 result
MeasureValueRankConfigurationDated
Vending-Bench 2: money after a yearUS dollars, higher is better $1,034 49 of 63 DeepSeek-V3.2-Exp 1 Oct 2026

Not shown: Not a real business; one simulated market with set rules.

Will DeepSeek-V3.2-Exp work for your product?

Benchmarks aren't your task. Paste the prompt your product uses and run it beside the models you could use instead, with every answer checked and costed.

New here? Your email starts a free trial. No credit card needed.