Model comparison

DeepSeek-V3.2-Speciale vs Gemini 3.1 Pro Preview

Gemini 3.1 Pro Preview is the stronger model overall, scoring 56.7 to 39.7 on the Noometry Index. DeepSeek-V3.2-Speciale costs 5.3× less per token, which makes it the better buy when Gemini 3.1 Pro Preview's lead doesn't matter for your workload.

Last verified October 10, 2026. 3 shared benchmarks.

DeepSeek-V3.2-Speciale

DeepSeek

39.7

Rank #162 Reported

Gemini 3.1 Pro Preview

Google

56.7

Rank #23 Confirmed

Summary

They share 3 benchmarks with published results for both. DeepSeek-V3.2-Speciale scores higher in 0 categories and Gemini 3.1 Pro Preview in 3 categories; 3 gaps are clear of the uncertainty.
The widest gap is in reasoning, where Gemini 3.1 Pro Preview leads 71.7 to 32.9.
The biggest single-benchmark swing is SimpleBench: 52.6% for DeepSeek-V3.2-Speciale and 79.6% for Gemini 3.1 Pro Preview.
DeepSeek-V3.2-Speciale is cheaper at $0.58 / $1.68 per million input/output tokens, against $2 / $12 for Gemini 3.1 Pro Preview.
Gemini 3.1 Pro Preview accepts more context: 1.05M tokens versus 128K.
DeepSeek-V3.2-Speciale has downloadable open weights; the other is API-only.

Side by side

DeepSeek-V3.2-Speciale and Gemini 3.1 Pro Preview specifications
	DeepSeek-V3.2-Speciale	Gemini 3.1 Pro Preview
Provider	DeepSeek	Google
Noometry Index	39.7	56.7
Released	2025-12-01	2026-02-19
Weights	Open	Proprietary
Context window	128K	1.05M
Max output	128K	66K
Input $ / M tokens	$0.58	$2
Output $ / M tokens	$1.68	$12
Results tracked	3	71

Sponsored placements are available on pages like this one. Advertise on Noometry

Category by category

Coding Gemini 3.1 Pro Preview leads

DeepSeek-V3.2-Speciale: 40.4 (#140), Gemini 3.1 Pro Preview: 42.5 (#99)

Coding benchmarks
Benchmark	DeepSeek-V3.2-Speciale	Gemini 3.1 Pro Preview
WeirdML	46.7%	72.1%
SWE-bench Verified	—	75.6%
DeepSWE	—	11.7%
LMArena WebDev	—	1447
SciCode	—	58.9%
GSO	—	22.6%
LMArena Coding	—	1484
MirrorCode	—	8.9%
ALE-Bench	—	1,161
AlgoTune	—	2.02

Agentic & Tool Use Not comparable

DeepSeek-V3.2-Speciale: —, Gemini 3.1 Pro Preview: 37.7 (#34)

Agentic & Tool Use benchmarks
Benchmark	DeepSeek-V3.2-Speciale	Gemini 3.1 Pro Preview
Terminal-Bench	—	80.2%
APEX-Agents	—	35.3%
τ²-bench Banking	—	26%
DeepResearch Bench	—	47.8%
PostTrainBench	—	22%
BALROG	—	57%
ExploitBench	—	26.1%
GBAEval	—	0.8%
GDP.pdf	—	17%
LMArena Search	—	1211
METR Time Horizons	—	77%
Vending-Bench 2	—	3,774

Reasoning Gemini 3.1 Pro Preview leads

DeepSeek-V3.2-Speciale: 32.9 (#73), Gemini 3.1 Pro Preview: 71.7 (#12)

Reasoning benchmarks
Benchmark	DeepSeek-V3.2-Speciale	Gemini 3.1 Pro Preview
SimpleBench	52.6%	79.6%
ARC-AGI-2	—	77.1%
NYT Connections (extended)	—	97.4%
ARC-AGI-1	—	98%
CritPt	—	17.7%
Chess Puzzles	—	55%
EnigmaEval	—	36.8%
Thematic Generalization	—	79.4%
EBR-Bench	—	14.3%
LMArena Hard Prompts	—	1485
Mystery Game Puzzles	—	34%
DTBench	—	97.1%
LMCA	—	53.8%
Epoch Capabilities Index	—	154.77
ForecastBench	—	59

Math Not comparable

DeepSeek-V3.2-Speciale: —, Gemini 3.1 Pro Preview: 62.1 (#34)

Math benchmarks
Benchmark	DeepSeek-V3.2-Speciale	Gemini 3.1 Pro Preview
FrontierMath (Tiers 1-3)	—	59.6%
FrontierMath Tier 4	—	26.8%
MathArena Final-Answer Competitions	—	86.5%
OTIS Mock AIME 2024-2025	—	95.6%
ProofBench	—	26%
LMArena Math	—	1485
FrontierMath (Feb 2025 set)	—	36.9%
FrontierMath Tier 4 (v1)	—	16.7%

Knowledge Not comparable

DeepSeek-V3.2-Speciale: —, Gemini 3.1 Pro Preview: 71.8 (#3)

Knowledge benchmarks
Benchmark	DeepSeek-V3.2-Speciale	Gemini 3.1 Pro Preview
GPQA Diamond	—	94.4%
Humanity's Last Exam	—	46.4%
SimpleQA Verified	—	73.5%
Vectara Hallucination Rate	—	10.4%
LMArena Expert	—	1485

Multimodal Not comparable

DeepSeek-V3.2-Speciale: —, Gemini 3.1 Pro Preview: 37.9 (#69)

Multimodal benchmarks
Benchmark	DeepSeek-V3.2-Speciale	Gemini 3.1 Pro Preview
LMArena Vision	—	1296
Blueprint-Bench 2	—	26.5%
Furniture Assembly	—	26.7%
LMArena Document	—	1444

Multilingual Not comparable

DeepSeek-V3.2-Speciale: —, Gemini 3.1 Pro Preview: 57.0 (#12)

Multilingual benchmarks
Benchmark	DeepSeek-V3.2-Speciale	Gemini 3.1 Pro Preview
LMArena Non-English	—	1477
LMArena Chinese	—	1529
LMArena French	—	1487
LMArena German	—	1491
LMArena Japanese	—	1493
LMArena Korean	—	1455
LMArena Russian	—	1498
LMArena Spanish	—	1479

Instruction Following Not comparable

DeepSeek-V3.2-Speciale: —, Gemini 3.1 Pro Preview: 77.0 (#32)

Instruction Following benchmarks
Benchmark	DeepSeek-V3.2-Speciale	Gemini 3.1 Pro Preview
LMArena Instruction Following	—	1466

Long Context Not comparable

DeepSeek-V3.2-Speciale: —, Gemini 3.1 Pro Preview: 47.4 (#18)

Long Context benchmarks
Benchmark	DeepSeek-V3.2-Speciale	Gemini 3.1 Pro Preview
CL-bench	—	20.8%
CL-bench Life	—	16.9%
LMArena Longer Query	—	1483

Writing & Preference Gemini 3.1 Pro Preview leads

DeepSeek-V3.2-Speciale: 46.0 (#222), Gemini 3.1 Pro Preview: 66.1 (#37)

Writing & Preference benchmarks
Benchmark	DeepSeek-V3.2-Speciale	Gemini 3.1 Pro Preview
EQ-Bench Creative Writing	1276	1491
LMArena Text	—	1481
LMArena Creative Writing	—	1482
EQ-Bench 4	—	1142
LMArena Multi-Turn	—	1488

Frequently asked questions

Is DeepSeek-V3.2-Speciale better than Gemini 3.1 Pro Preview?

Which is cheaper, DeepSeek-V3.2-Speciale or Gemini 3.1 Pro Preview?

DeepSeek-V3.2-Speciale is cheaper. It lists at $0.58 per million input tokens and $1.68 per million output tokens; Gemini 3.1 Pro Preview lists at $2 and $12.

Is DeepSeek-V3.2-Speciale or Gemini 3.1 Pro Preview better for coding?

Gemini 3.1 Pro Preview scores higher on coding benchmarks: 42.5 versus 40.4 in the Noometry coding category.

Which has the bigger context window?

Gemini 3.1 Pro Preview does, with 1.05M tokens against 128K.

How many benchmarks do DeepSeek-V3.2-Speciale and Gemini 3.1 Pro Preview share?

3 benchmarks have published results for both models. DeepSeek-V3.2-Speciale has 3 scored results on Noometry and Gemini 3.1 Pro Preview has 71.