Grok 4.20

Grok 4.20 ranks first on 2 of the 117 benchmarks with five or more models scored.

xAI Released February 17th, 2026 Compare with others

121

Benchmarks with a score

Ranked first, out of five or more models

In the top three

xAI Feb 2026

Grok 4.20

Closed weights

152.3 #60 of 256

Epoch Capabilities Index

Terminal-Bench 2.0 57.3% #14 /43
ZeroBench 10% #39 /76
GPQA Diamond 89.3% #55 /229

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

Benchmark	Score	Rank	Source
BullshitBench	61.8%	#16 of 107	Official leaderboard
CL-bench	22.2%	#4 of 24	Official leaderboard
StudyArena Text	1,250	#23 of 85	Official leaderboard
Extended NYT Connections	85.4%	#33 of 103	Official leaderboard
BullshitBench V2	56%	#43 of 129	Official leaderboard
Longform Creative Writing (EQ-Bench)	68.5	#45 of 119	Official leaderboard
Creative Writing v3	65.7	#69 of 120	Official leaderboard
CL-bench Life	11.9%	#11 of 16	Official leaderboard
SimpleQA Verified	30.2%	#72 of 94	Epoch AI run
EQ-Bench 3	893	#69 of 78	Official leaderboard
OpenAI MRCR v2	15.9%	#42 of 47	Official leaderboard

Math and science

Benchmark	Score	Rank	Source
GPQA Diamond	89.3%	#55 of 229	Epoch AI run
DTBench	90.1%	#49 of 180	Official leaderboard
OTIS Mock AIME 2024-2025	92.2%	#61 of 212	Epoch AI run
ARC-AGI-2	65.1%	#31 of 98	Official leaderboard
ForecastBench	0.13	#29 of 82	Official leaderboard
ARC-AGI	89.5%	#36 of 95	Official leaderboard
Echo General AI Prediction Leaderboard	1,001	#9 of 17	Official leaderboard
ProofBench	14%	#39 of 59	Official leaderboard
FrontierMath Tiers 1-3 v2	44.9%	#61 of 92	Epoch AI run
FrontierMath Tier 4 v2	17.1%	#55 of 74	Epoch AI run
Leipzig Benchmark	5.00%	#19 of 19	Official leaderboard
Human Behavior Bench	73.5	#11 of 11	Official leaderboard

Coding

Benchmark	Score	Rank	Source
IOI (Vals)	30.2%	#16 of 55	Official leaderboard
DesignArena Agentic Overall	1,086	#52 of 74	Official leaderboard
WebDev Arena	1,374	#75 of 105	arena.ai, via Epoch AI
DesignArena Frontend Web App	1,126	#38 of 53	Official leaderboard
AutoLab	0.35	#8 of 10	Official leaderboard
DesignArena Full Stack	1,033	#40 of 49	Official leaderboard
DesignArena Android (Kotlin)	1,036	#41 of 46	Official leaderboard
Vibe Code Bench	4.06%	#107 of 120	Official leaderboard
DesignArena React Native	1,084	#51 of 56	Official leaderboard
Code Migration	0.31%	#77 of 78	Official leaderboard

Agents and tools

Benchmark	Score	Rank	Source
LisanBench	3856.3	#31 of 133	Official leaderboard
LLM Chess	843	#29 of 123	Official leaderboard
Game Arena Bargaining	220	#10 of 39	Official leaderboard
PinchBench 2.0	80.3%	#17 of 57	Official leaderboard
Terminal-Bench 2.0	57.3%	#14 of 43	Official leaderboard
Game Arena Five in a Row	618	#8 of 20	Official leaderboard
Pencil Puzzle Bench	6.70%	#22 of 53	Official leaderboard
PinchBench	71.2%	#21 of 48	Official leaderboard
Chess Puzzles	29%	#53 of 93	Epoch AI run
Game Arena Chess	666	#28 of 42	Official leaderboard
Game Arena Four in a Row	290	#29 of 43	Official leaderboard
Game Arena Coin Game	249	#29 of 39	Official leaderboard
Game Arena Lines of Action	284	#29 of 39	Official leaderboard
Game Arena Dots and Boxes	629	#30 of 40	Official leaderboard
Game Arena Ultimate Tic-Tac-Toe	167	#30 of 39	Official leaderboard
Kaggle Game Arena	194	#33 of 42	Official leaderboard
TERMS-Bench	0.60	#9 of 11	Official leaderboard
Game Arena Nine Men's Morris	378	#32 of 39	Official leaderboard
Game Arena Clobber	110	#33 of 39	Official leaderboard
Game Arena Go	471	#33 of 39	Official leaderboard
Game Arena Word Association	58	#35 of 40	Official leaderboard
Game Arena Checkers	417	#36 of 41	Official leaderboard
CEO-Bench	$0	#17 of 19	Official leaderboard
Game Arena Reversi	122	#35 of 39	Official leaderboard
Game Arena Poker	-25.68	#29 of 31	Official leaderboard
Game Arena Word Art	67	#37 of 39	Official leaderboard
Game Arena Dark Hex	44	#39 of 41	Official leaderboard
ClawProBench	43%	#47 of 49	Official leaderboard
ARC-AGI-3	0.09%	#27 of 27	Official leaderboard
Long-Horizon-Terminal-Bench (LHTB)	10.2%	#25 of 25	Official leaderboard

Professional work

Benchmark	Score	Rank	Source
TaxEval v2	74.1%	#40 of 129	Official leaderboard
CorpFin v2	63.7%	#44 of 119	Official leaderboard
prinzbench	43.0	#11 of 26	Official leaderboard
Vals Finance Agent	52.3%	#24 of 49	Official leaderboard
Vending-Bench 2	$4,663	#38 of 67	Official leaderboard
ProfBench	43.8%	#51 of 72	Official leaderboard
CaseLaw v2	54.5%	#38 of 53	Official leaderboard
LegalBench	77.7%	#99 of 138	Official leaderboard
FoodTruck Bench	$1,338	#46 of 59	Official leaderboard
Vals Index v1	46.1%	#28 of 34	Official leaderboard
Harvey LAB (Legal Agent Benchmark)	0%	#66 of 80	Official leaderboard
Legal Research Bench	13.9%	#79 of 93	Official leaderboard
MedCode	32.2%	#84 of 96	Official leaderboard
MortgageTax	45.4%	#78 of 87	Official leaderboard
Tax Agent Bench	6.95%	#76 of 84	Official leaderboard
Vals Finance Agent v2	28.5%	#74 of 80	Official leaderboard
MedScribe	63.4%	#91 of 97	Official leaderboard
TRLawBench	7.38	#32 of 34	Official leaderboard
Vals Index v2	17.6%	#62 of 65	Official leaderboard
Excel Modeling Benchmark	11.7%	#73 of 76	Official leaderboard
PhysicianBench	5.30%	#13 of 13	Official leaderboard
HealthBench Professional	36.1%	#6 of 6	Official leaderboard

Vision, audio and robotics

Benchmark	Score	Rank	Source
ZeroBench	10%	#39 of 76	Official leaderboard
MineBench	1,280	#50 of 75	Official leaderboard
Blueprint-Bench 2	0%	#36 of 38	Official leaderboard
ArtifactArena	745	#21 of 21	Official leaderboard

Safety and security

Benchmark	Score	Rank	Source
WMDP Biology	91%	#1 of 8	Lab model card
WMDP Chemistry	90%	#1 of 7	Lab model card
LAB-Bench CloningScenarios	67%	#3 of 12	Lab model card
SycophancyEval (answer sycophancy)	4.00%	#1 of 4	Lab model card
BioTIER Permit	99.7%	#22 of 69	Official leaderboard
PCT (Polarized Contrastive Pairs)	39.9%	#15 of 46	Official leaderboard
LAB-Bench FigQA	66%	#4 of 11	Lab model card
Human Pathogen Capabilities Test v2 (HPCT v2)	56.2%	#32 of 71	Official leaderboard
Virology Capabilities Test (VCT)	54%	#14 of 31	Lab model card
BioTIER Refuse	66.9%	#33 of 69	Official leaderboard
General refusals (xAI)	0%	#4 of 8	Lab model card
WMDP Cyber	91%	#1 of 2	Lab model card
Cybench	53%	#10 of 18	Lab model card
nocot-bench (NCRI)	67.1	#111 of 185	Official leaderboard
Refusals + System Jailbreak (xAI)	0%	#3 of 5	Lab model card
LAB-Bench ProtocolQA	79%	#10 of 16	Lab model card
MASK	73%	#14 of 19	Lab model card
ProtocolQA Open-Ended (xAI)	79%	#3 of 4	Lab model card
Refusals + User Jailbreak (xAI)	1.00%	#4 of 5	Lab model card
Dictatorship Eval	25.2%	#17 of 20	Official leaderboard
MakeMeSay	8.00%	#9 of 10	Lab model card
AgentDojo	33%	#8 of 8	Lab model card
AgentHarm	30%	#6 of 6	Lab model card
xAI sycophancy eval	35%	#4 of 4	Lab model card

Other

Benchmark	Score	Rank	Source
Conceptual Reasoning Index	53.7	#56 of 151	Official leaderboard
ACCoRD	68.8	#57 of 152	Official leaderboard
LMCA	38.7%	#61 of 147	Official leaderboard
CAIS Risk Index	51.7	#35 of 67	Official leaderboard
CAIS Vision Capabilities Index	48.5	#28 of 53	Official leaderboard
CAIS Text Capabilities Index	24.3	#46 of 68	Official leaderboard
SAGE (Vals)	38.2%	#60 of 81	Official leaderboard
Vals Multimodal Index	39.1%	#32 of 33	Official leaderboard

xAI's models over time

xAI's models in color, against every other model on the Epoch Capabilities Index, by release date.

Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

Grok 4.20 model documentation Mar 9, 2026

Its system card

Every result in the longer documents the lab published with it. 1 of them isn't in the launch post.

Grok 4.20 System Card Apr 7, 2026

Hallucination

HLE Calibration 19

Safety

Robustness

AgentDojo 33

Alignment