xAIReasoning model

Grok 3 mini

Name: Grok 3 mini benchmark results
Creator: BenchLeader

Best configuration ranks #227 of 610 on the BenchLeader Index at 53.3 ±4.1 (high reasoning effort). Last measured 2 Sept 2026. Released 9 Apr 2025.

Blended price: $1.56/M; $1.25 in · $2.50 out
Output speed: –
First answer: –
Context: 131k

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

Overall index53
Reasoning55
Coding49
Maths61
Human preference54

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	50.2	#291	–	–	$0.0013	Maths 59 · Reasoning 42
medium	–	–	–	–	$0.0013	Long context 56
highbest	53.3	#227	–	–	$0.0013	Coding 49 · Human preference 54 · Maths 61 · Reasoning 55
thinking	51.0	#274	56 tok/s	37 s	$0.0003	Agents & tools 49 · Composite 47 · Instruction following 47 · Knowledge 61 · Long context 54
default	52.0	#261	–	–	$0.0013	Coding 52 · Human preference 53 · Reasoning 55

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	medium	high	thinking	default	Source
GPQA Diamond	76.3%#129	–	75.5%#136	–	–	Epoch AI Benchmarking Hub
LMArena Hard Prompts	–	–	1378#169	–	1370#175	LMArena
GPQA Diamond (AA)not in index	–	–	–	79.1%#196	–	Artificial Analysis
Humanity's Last Exam (AA)not in index	–	–	–	11.0%#239	–	Artificial Analysis
Kagi LLM Benchmark	–	–	–	–	61.3%#47	Kagi LLM Benchmark
ARC-AGI-1	16.5%#161	–	–	–	–	ARC Prize
ARC-AGI-2	0.4%#171	–	–	–	–	ARC Prize

Coding

Benchmark	low	medium	high	thinking	default	Source
WeirdML	–	–	42.6%#91	–	–	WeirdML
LMArena Coding	–	–	1393#178	–	1386#186	LMArena
Aider Polyglot	–	–	–	–	49.3%#25	Aider polyglot leaderboard

Agents & tools

Benchmark	low	medium	high	thinking	default	Source
Terminal-Bench Hard	–	–	–	17.4%#173	–	Artificial Analysis
τ²-Bench Telecom (AA)not in index	–	–	–	90.3%#65	–	Artificial Analysis

Maths

Benchmark	low	medium	high	thinking	default	Source
OTIS Mock AIME	62.2%#143	–	77.8%#107	–	–	Epoch AI Benchmarking Hub
MATH Level 5	90.9%#19	–	88.1%#23	–	–	Epoch AI Benchmarking Hub

Knowledge

Benchmark	low	medium	high	thinking	default	Source
AA-Omniscience	–	–	–	-6.8#121	–	Artificial Analysis

Instruction following

Benchmark	low	medium	high	thinking	default	Source
IFBench	–	–	–	45.9%#195	–	Artificial Analysis

Human preference

Benchmark	low	medium	high	thinking	default	Source
LMArena Text	–	–	1364#165	–	1358#171	LMArena

Long context

Benchmark	low	medium	high	thinking	default	Source
Fiction.LiveBench 120k	–	65.6%#10	–	–	–	Fiction.live
AA-LCR	–	–	–	55.3%#233	–	Artificial Analysis

Composite

Benchmark	low	medium	high	thinking	default	Source
Epoch Capabilities Indexnot in index	–	–	–	–	140.4#96	Epoch AI Benchmarking Hub
AA Intelligence Index	–	–	–	14.6#227	–	Artificial Analysis

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	$0.0013	–	–
Summarise a 30-page report	12,000 / 600	$0.017	–	–
Code edit	6,000 / 1,500	$0.011	–	–
Agentic coding session	60,000 / 4,000	$0.085	–	–
Structured extraction	2,000 / 200	$0.0030	–	–