GoogleReasoning model

Gemini 3 Pro

Name: Gemini 3 Pro benchmark results
Creator: BenchLeader

Best configuration ranks #64 of 610 on the BenchLeader Index at 61.6 ±3.1. Last measured 8 Sept 2026. Released 18 Nov 2025.

Blended price: $4.50/M; $2.00 in · $12.00 out
Output speed: –
First answer: –
Context: 1.0M

How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.

Overall index62
Reasoning67
Coding59
Agents & tools62
Maths48
Knowledge61
Instruction following64
Human preference69
Multimodal66
Long context65
Composite64

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. “Default” means the publisher did not say which setting was used.

Effort	Index	Rank	Speed	First answer	Chat reply cost	Categories
low	56.2	#161	–	–	$0.0044	Agents & tools 63 · Composite 57 · Instruction following 51 · Knowledge 65 · Long context 63
high	57.0	#151	–	–	$0.0044	Agents & tools 54 · Coding 61 · Knowledge 60 · Maths 63
defaultbest	61.6	#64	–	–	$0.0044	Agents & tools 62 · Coding 59 · Composite 64 · Human preference 69 · Instruction following 64 · Knowledge 61 · Long context 65 · Maths 48 · Multimodal 66 · Reasoning 67

Benchmark results

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.

Reasoning

Benchmark	low	high	default	Source
GPQA Diamond	–	–	92.6%#21	Epoch AI Benchmarking Hub
Humanity's Last Exam	–	–	37.5%#5	Scale AI / CAIS
SimpleBench	–	–	76.4%#7	SimpleBench
LMArena Hard Prompts	–	–	1504#20	LMArena
GPQA Diamond (AA)not in index	88.7%#86	–	90.8%#55	Artificial Analysis
Humanity's Last Exam (AA)not in index	29.5%#108	–	39.7%#57	Artificial Analysis
GPQA Diamond (Vals)not in index	–	91.7%#18	–	Vals AI
Kagi LLM Benchmark	–	–	80.1%#9	Kagi LLM Benchmark
ARC-AGI-1	–	–	75.0%#92	ARC Prize
ARC-AGI-2	–	–	54.0%#68	ARC Prize

Coding

Benchmark	low	high	default	Source
SWE-bench Verified (Epoch)	–	–	72.9%#21	Epoch AI Benchmarking Hub
WeirdML	–	–	69.9%#31	WeirdML
GSO-Bench	–	–	18.6%#13	GSO-Bench
LMArena Coding	–	–	1518#32	LMArena
LMArena WebDev	–	–	1439#57	LMArena
LiveCodeBench	–	–	86.4%#22	Vals AI
IOI	–	38.8%#14	–	Vals AI
SWE-bench (Vals)not in index	–	76.4%#35	–	Vals AI
SWE-Bench Pro	–	–	43.3%#8	Scale AI SEAL
SWE-bench Verified (bash only)	–	69.6%#13	74.2%#6	SWE-bench
SWE-bench Verified (any scaffold)not in index	–	–	77.4%#3	SWE-bench

Agents & tools

Benchmark	low	high	default	Source
Terminal-Bench	–	–	69.4%#7	Terminal-Bench
GDPval	–	–	40.3%#5	OpenAI
Remote Labor Index	–	–	1.3%#11	Scale AI / CAIS
APEX-Agents	–	–	31.5%#27	Mercor
Terminal-Bench Hard	34.1%#87	–	41.7%#48	Artificial Analysis
τ²-Bench Telecom (AA)not in index	68.1%#157	–	87.1%#79	Artificial Analysis
MCP Atlas	–	–	70.3%#18	Scale AI SEAL
BFCL Overall	–	–	72.5%#3	Berkeley Function Calling Leaderboard
τ²-bench	–	82.5%#5	–	τ²-bench

Maths

Benchmark	low	high	default	Source
OTIS Mock AIME	–	–	91.4%#53	Epoch AI Benchmarking Hub
ProofBench	–	–	20.0%#36	Vals AI
AIME (Vals)	–	96.7%#4	–	Vals AI
MGSM	–	93.9%#7	–	Vals AI
AIME 2026	–	–	91.7%#24	MathArena
HMMT February 2026	–	–	86.4%#20	MathArena
MathArena Apex	–	–	23.4%#14	MathArena

Knowledge

Benchmark	low	high	default	Source
AA-Omniscience	1.8#83	–	15.3#51	Artificial Analysis
MMLU-Pro	–	–	90.1%#7	Vals AI
LegalBench	–	87.0%#5	–	Vals AI
CorpFin	–	63.7%#42	–	Vals AI
TaxEval	–	72.6%#61	–	Vals AI
MedQA	–	96.0%#8	–	Vals AI
PRBench Finance	–	–	39.2%#24	Scale AI SEAL
PRBench Legal	–	–	40.6%#23	Scale AI SEAL
MultiNRC	–	–	59.0%#6	Scale AI SEAL

Instruction following

Benchmark	low	high	default	Source
IFBench	49.7%#174	–	70.4%#67	Artificial Analysis
MultiChallenge	–	–	65.7%#5	Scale AI SEAL
TutorBench	–	–	53.7%#9	Scale AI SEAL

Human preference

Benchmark	low	high	default	Source	Trend
LMArena Text	–	–	1486#16	LMArena

Multimodal

Benchmark	low	high	default	Source
LMArena Vision	–	–	1305#11	LMArena
MMMU-Pro	–	–	80.2%#35	Artificial Analysis
VISTA	–	–	51.5%#7	Scale AI SEAL
MMMU-Pro (official)not in index	–	–	81.0%#1	MMMU

Long context

Benchmark	low	high	default	Source	Trend
AA-LCR	74.0%#120	–	76.0%#102	Artificial Analysis

Composite

Benchmark	low	high	default	Source	Trend
Epoch Capabilities Indexnot in index	–	–	153.0#33	Epoch AI Benchmarking Hub
AA Intelligence Index	22.3#147	–	28.0#94	Artificial Analysis

Where it wins

Benchmarks where this configuration ranks in the top five of every configuration measured.

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. Reasoning tokens are not modelled.

Workload	Tokens in / out	Cost	With caching	Time
Chat reply	400 / 300	$0.0044	–	–
Summarise a 30-page report	12,000 / 600	$0.031	–	–
Code edit	6,000 / 1,500	$0.030	–	–
Agentic coding session	60,000 / 4,000	$0.168	–	–
Structured extraction	2,000 / 200	$0.0064	–	–