western lab · United States

xAI

Vertically integrated lab operating the Colossus training cluster in Memphis, shipping the Grok model family with tight distribution through X.

Founders

Elon Musk

Model Benchmarks

Model	Released	Intelligence	Coding	GPQA	HLE	SciCode	Terminal-Bench Hard	Tau2	LCR	IFBench	Speed	TTFT	$/1M in	$/1M out	$/1M blended
Grok 4.5 (high)	2026-07-08	53.8	72.4	93.1%	40.3%	54.1%	—	—	67.7%	—	67.5 t/s	14.2s	$2	$6	$3
Grok Build 0.1 0616	2026-06-16	39.8	51.5	89.5%	36%	50.2%	—	—	64.7%	—	0 t/s	0s	$1	$2	$1.25
Grok 4.3 (high)	2026-04-30	37.6	42.2	90.1%	35%	47.3%	37.9%	97.7%	64.3%	81.3%	0 t/s	0s	$1.25	$2.5	$1.56
Grok 4.20 0309 v2 (Reasoning)	2026-04-07	37	—	91.1%	32.2%	45.6%	37.9%	93%	58%	81.2%	0 t/s	0s	$1.25	$2.5	$1.56
Grok 4.20 0309 (Reasoning)	2026-03-10	36.5	—	88.5%	30%	44.7%	40.9%	96.5%	59%	82.9%	0 t/s	0s	$2	$6	$3
Grok 4.3 (medium)	2026-04-30	36	—	89%	28.1%	44.6%	30.3%	91.2%	65%	83.3%	141.3 t/s	18.48s	$1.25	$2.5	$1.56
Grok 4.3 (low)	2026-04-30	35.4	—	84.3%	17.3%	41.9%	26.5%	88.9%	64%	81%	132 t/s	6.62s	$1.25	$2.5	$1.56
Grok 4	2025-07-10	33.3	—	87.7%	23.9%	45.7%	37.9%	74.9%	68%	53.7%	0 t/s	0s	$3	$15	$6
Grok 4.1 Fast (Reasoning)	2025-11-19	30.6	—	85.3%	17.6%	44.2%	24.2%	93.3%	68%	52.7%	0 t/s	0s	$0	$0	$0
Grok 4 Fast (Reasoning)	2025-09-19	27.4	—	84.7%	17%	44.2%	18.9%	65.8%	64.7%	50.5%	0 t/s	0s	$0.2	$0.5	$0.28

Indices on Artificial Analysis 0-100 scale · eval scores are raw accuracy · “—” means the metric is not published for that model

disclosed

Funding rounds

Partners

Publicly announced rounds and partnerships only · as of Aug 2026