Model comparison

DeepSeek-V3 vs Gemini 2.5 Pro

Gemini 2.5 Pro is the stronger model overall, scoring 44.9 to 39.2 on the Noometry Index. DeepSeek-V3 costs 8.5× less per token, which makes it the better buy when Gemini 2.5 Pro's lead doesn't matter for your workload.

Last verified October 10, 2026. 49 shared benchmarks.

DeepSeek-V3

DeepSeek

39.2

Rank #157 Confirmed

Gemini 2.5 Pro

Google

44.9

Rank #67 Confirmed

Summary

They share 49 benchmarks with published results for both. DeepSeek-V3 scores higher in 0 categories and Gemini 2.5 Pro in 8 categories; 7 gaps are clear of the uncertainty.
The widest gap is in long context, where Gemini 2.5 Pro leads 59.7 to 34.0.
The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 37.8% for DeepSeek-V3 and 84.7% for Gemini 2.5 Pro.
DeepSeek-V3 is cheaper at $0.24 / $0.90 per million input/output tokens, against $1.25 / $10 for Gemini 2.5 Pro.
Gemini 2.5 Pro accepts more context: 1.05M tokens versus 164K.
DeepSeek-V3 has downloadable open weights; the other is API-only.

Side by side

DeepSeek-V3 and Gemini 2.5 Pro specifications
	DeepSeek-V3	Gemini 2.5 Pro
Provider	DeepSeek	Google
Noometry Index	39.2	44.9
Released	2024-12-26	2025-03-25
Weights	Open	Proprietary
Context window	164K	1.05M
Max output	164K	66K
Input $ / M tokens	$0.24	$1.25
Output $ / M tokens	$0.90	$10
Results tracked	60	78

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

Category by category

Coding Too close to call

DeepSeek-V3: 41.9 (#99), Gemini 2.5 Pro: 42.8 (#86)

Coding benchmarks
Benchmark	DeepSeek-V3	Gemini 2.5 Pro
Aider Polyglot	55.1%	83.1%
SciCode	35.8%	42.8%
WeirdML	36.1%	54%
LiveBench Coding	70.9%	85.9%
LMArena Coding	1368	1452
SWE-bench Verified	—	57.6%
SWE-bench Verified (bash only)	—	53.6%
LMArena WebDev	—	1227
GSO	—	3.9%
BigCodeBench Instruct	50%	—
BigCodeBench Complete	62.2%	—
CadEval	—	64%
ALE-Bench	—	785.52
AlgoTune	—	1.51
HumanEval+	86.6%	—
MBPP+	73%	—

Agentic & Tool Use Not comparable

DeepSeek-V3: —, Gemini 2.5 Pro: 29.1 (#81)

Agentic & Tool Use benchmarks
Benchmark	DeepSeek-V3	Gemini 2.5 Pro
METR Time Horizons	49.6%	55.4%
Terminal-Bench	—	32.6%
GDPval	—	23.3%
Remote Labor Index	—	0.8%
TheAgentCompany	—	30.3%
τ²-bench Banking	—	13.7%
DeepResearch Bench	—	42.8%
BALROG	—	43.3%
LMArena Search	—	1142
Vending-Bench 2	—	573.64

Reasoning Gemini 2.5 Pro leads

DeepSeek-V3: 20.0 (#226), Gemini 2.5 Pro: 28.8 (#85)

Reasoning benchmarks
Benchmark	DeepSeek-V3	Gemini 2.5 Pro
SimpleBench	27.2%	62.4%
Kagi LLM Benchmark	52.3%	70.3%
CritPt	0%	2%
LiveBench Reasoning	65.8%	89.8%
LMArena Hard Prompts	1365	1455
DTBench	64.8%	82.4%
LiveBench Data Analysis	60.9%	79.9%
LMCA	15.5%	34.8%
Epoch Capabilities Index	135.95	145.32
ForecastBench	59.1	61.3
LiveBench	66.9%	82.3%
ARC-AGI-2	—	4.9%
ARC-AGI-1	—	41%
Chess Puzzles	—	20%
EnigmaEval	—	5.6%
BIG-Bench Hard	87.5%	—
HellaSwag	88.9%	—
PIQA	84.7%	—
WinoGrande	85.2%	—

Math Gemini 2.5 Pro leads

DeepSeek-V3: 30.4 (#208), Gemini 2.5 Pro: 31.9 (#191)

Math benchmarks
Benchmark	DeepSeek-V3	Gemini 2.5 Pro
OTIS Mock AIME 2024-2025	37.8%	84.7%
Omni-MATH	40.3%	41.6%
LiveBench Math	73.5%	90.2%
LMArena Math	1373	1450
MATH Level 5	75.5%	95.9%
FrontierMath (Feb 2025 set)	1.7%	14.1%
FrontierMath (Tiers 1-3)	—	24.6%
FrontierMath Tier 4	—	0%
FrontierMath Tier 4 (v1)	—	4.2%

Knowledge Gemini 2.5 Pro leads

DeepSeek-V3: 37.5 (#147), Gemini 2.5 Pro: 55.8 (#41)

Knowledge benchmarks
Benchmark	DeepSeek-V3	Gemini 2.5 Pro
GPQA Diamond	67.6%	85.3%
MMLU-Pro	72.3%	86.3%
Confabulations	26.1%	10.6%
Vectara Hallucination Rate	6.1%	7%
GPQA (HELM)	53.8%	74.9%
LMArena Expert	1352	1452
Humanity's Last Exam	—	21.6%
ARC (AI2) Challenge	95.3%	—
MMLU	87.2%	—
TriviaQA	82.9%	—

Multimodal Not comparable

DeepSeek-V3: —, Gemini 2.5 Pro: 44.5 (#16)

Multimodal benchmarks
Benchmark	DeepSeek-V3	Gemini 2.5 Pro
LMArena Vision	—	1263
GeoBench	—	86%
VPCT	—	48%
LMArena Document	—	1421
SpatialViz-Bench	—	44.7%

Multilingual Gemini 2.5 Pro leads

DeepSeek-V3: 48.8 (#138), Gemini 2.5 Pro: 55.5 (#30)

Multilingual benchmarks
Benchmark	DeepSeek-V3	Gemini 2.5 Pro
LMArena Non-English	1358	1452
LMArena Chinese	1390	1507
LMArena French	1386	1473
LMArena German	1374	1487
LMArena Japanese	1333	1462
LMArena Korean	1321	1435
LMArena Russian	1372	1460
LMArena Spanish	1358	1473

Instruction Following Gemini 2.5 Pro leads

DeepSeek-V3: 73.0 (#125), Gemini 2.5 Pro: 75.2 (#71)

Instruction Following benchmarks
Benchmark	DeepSeek-V3	Gemini 2.5 Pro
LiveBench Instruction Following	81.5%	80.6%
IFEval	83.2%	84%
LMArena Instruction Following	1345	1436

Long Context Gemini 2.5 Pro leads

DeepSeek-V3: 34.0 (#248), Gemini 2.5 Pro: 59.7 (#5)

Long Context benchmarks
Benchmark	DeepSeek-V3	Gemini 2.5 Pro
Fiction.LiveBench	50%	91.7%
LMArena Longer Query	1353	1449

Writing & Preference Gemini 2.5 Pro leads

DeepSeek-V3: 57.6 (#124), Gemini 2.5 Pro: 63.9 (#60)

Writing & Preference benchmarks
Benchmark	DeepSeek-V3	Gemini 2.5 Pro
LMArena Text	1375	1458
LMArena Creative Writing	1364	1453
Short-Story Creative Writing	77%	83.8%
EQ-Bench Creative Writing	1472	1421
WildBench	83%	85.7%
LMArena Multi-Turn	1389	1453
LiveBench Language	49.1%	67.8%

Frequently asked questions

Is DeepSeek-V3 better than Gemini 2.5 Pro?

Which is cheaper, DeepSeek-V3 or Gemini 2.5 Pro?

DeepSeek-V3 is cheaper. It lists at $0.24 per million input tokens and $0.90 per million output tokens; Gemini 2.5 Pro lists at $1.25 and $10.

Is DeepSeek-V3 or Gemini 2.5 Pro better for coding?

They score almost the same on coding (41.9 vs 42.8); test both on your own repository before choosing.

Which has the bigger context window?

Gemini 2.5 Pro does, with 1.05M tokens against 164K.

How many benchmarks do DeepSeek-V3 and Gemini 2.5 Pro share?

49 benchmarks have published results for both models. DeepSeek-V3 has 60 scored results on Noometry and Gemini 2.5 Pro has 78.