SpaceXAIReasoning model

Grok 4.6

Reasoning effort

Grok 4.6 is a SpaceXAI proprietary reasoning model, released 11 Aug 2026. Its best configuration (medium reasoning effort) ranks #53 of 756 on the BenchLeader Index at 64.3 ±4.6, in the upper half. The ± is the point: 129 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in composite (79) and lowest in coding (61). At $3.00 per million tokens blended it is pricier than most ranked models. Output speed of 76 tokens per second puts it slower than most, with a first answer in 24.1 s. It has been measured at 5 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 12 Aug 2026.

Blended price: $3.00/M; $2.00 in · $6.00 out
Output speed: 76 tok/s; measured by Artificial Analysis
First answer: 24 s; first token 2.48 s
Context: 500k
Full answer: 31 s; median, reasoning included
Cost per run: $1.50; one full Intelligence Index run
Released: 11 Aug 2026

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

Overall index64
Reasoning65
Coding61
Agents & tools74
Knowledge74
Long context66
Composite79

Versions

SpaceXAI has shipped 9 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.

Model	Released	Index	Rank
Grok 4.7high	21 Sept 2026	63.6	#61
Grok 4.6mediumthis page	11 Aug 2026	64.3	#53
Grok 4.5high	16 Jul 2026	59.6	#119
Grok 4.3medium	17 Apr 2026	58.8	#135
Grok 4.20thinking	5 Mar 2026	59.1	#127
Grok 4.1	19 Nov 2025	57.5	#166
Grok 4	9 Jul 2025	58.1	#152
Grok 3	17 Feb 2025	51.6	#325
Grok 2	12 Dec 2024	42.0	#585

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. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.

Effort	Index	Rank	Speed	First answer	Chat reply cost	Categories
low	59.5	#122	68 tok/s	7.44 s	$0.0026	Agents & tools 67 · Coding 53 · Composite 70 · Knowledge 74 · Long context 66 · Reasoning 50
mediumbest	64.3	#53	76 tok/s	24 s	$0.0026	Agents & tools 74 · Coding 61 · Composite 79 · Knowledge 74 · Long context 66 · Reasoning 65
high	63.3	#66	77 tok/s	30 s	$0.0026	Agents & tools 58 · Coding 62 · Composite 80 · Human preference 64 · Knowledge 63 · Long context 66 · Maths 64 · Multimodal 61 · Reasoning 66
xhigh	62.7	#77	72 tok/s	30 s	$0.0026	Agents & tools 65 · Coding 57 · Composite 80 · Knowledge 66 · Long context 66 · Maths 59 · Reasoning 64
not stated	57.9	#158	59 tok/s	2.48 s	$0.0026	Agents & tools 65 · Coding 63 · Composite 59 · Maths 55 · Reasoning 72

What more thinking costs

Turning the effort up buys index points and multiplies the bill. Here xhigh costs 4.9× low for +3.2 on the index — 1.4 points per doubling of spend. Compare that with other models.

How its index has moved

8 Sept 2026 to 5 Oct 2026

medium
high
xhigh
low

The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.

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	xhigh	not stated	Source
GPQA Diamond	–	–	94.0%#11	93.2%#19	–	Epoch AI Benchmarking Hub
SimpleBench	–	–	–	–	75.9%#8	SimpleBench
LMArena Hard Prompts	–	–	1479#67	–	–	LMArena
LiveBench Reasoningnot in index	–	–	–	–	90.5%#12	LiveBench
GPQA Diamond (AA)not in index	87.9%#90	93.5%#14	95.0%#4	93.5%#14	–	Artificial Analysis
Humanity's Last Exam (AA)not in index	27.6%#157	42.1%#66	42.9%#55	44.1%#49	–	Artificial Analysis
GPQA Diamond (Vals)not in index	–	–	94.7%#3	–	–	Vals AI
ARC-AGI-1	74.8%#113	87.5%#76	87.0%#81	87.0%#81	–	ARC Prize
ARC-AGI-2	27.6%#117	61.3%#72	65.1%#65	67.1%#60	–	ARC Prize
ARC-AGI-3	–	–	–	2.1%#23	–	ARC Prize
CritPt	5.7%#129	17.7%#65	17.1%#72	19.7%#58	–	Artificial Analysis
MysteryMechanismnot in index	–	–	30.6%#11	–	–	Vals AI
Chess Puzzlesnot in index	–	–	40.0%#24	31.0%#48	–	Epoch AI Benchmarking Hub
Mystery Game Puzzlesnot in index	–	–	–	34.0%#21	–	Epoch AI Benchmarking Hub
LMCAnot in index	–	–	–	48.5%#61	–	Epoch AI Benchmarking Hub
DTBenchnot in index	–	–	–	97.3%#8	–	Epoch AI Benchmarking Hub

Coding

Benchmark	low	medium	high	xhigh	not stated	Source
SciCode	48.4%#111	54.6%#52	56.5%#32	51.6%#82	–	SciCode
WeirdML	–	–	67.3%#37	–	–	WeirdML
FrontierCode	–	–	–	–	48.0%#10	Cognition
LMArena Coding	–	–	1507#64	–	–	LMArena
LMArena WebDev	–	–	1620#22	–	–	LMArena
LiveBench Codingnot in index	–	–	–	–	76.8%#41	LiveBench
SciCode (AA)not in index	49.4%#106	55.9%#39	56.5%#33	53.0%#73	–	Artificial Analysis
LiveCodeBench	–	–	88.2%#7	–	–	Vals AI
IOI	–	–	47.6%#25	–	–	Vals AI
SWE-bench (Vals)not in index	–	–	95.6%#4	–	–	Vals AI
Code Migrationnot in index	–	–	44.6%#17	–	–	Vals AI
Vibe Code Bench 1-100not in index	–	–	14.8%#15	–	–	Vals AI
Vibe Code Bench v1.1not in index	–	–	76.2%#27	–	–	Vals AI
DeepSWEnot in index	41.6%#55	67.5%#20	65.2%#28	66.7%#24	–	Epoch AI Benchmarking Hub
CursorBenchnot in index	33.4%#41	36.1%#35	40.4%#27	41.4%#24	–	Cursor
ALE-Benchnot in index	–	–	–	1508.1#19	–	Epoch AI Benchmarking Hub
FrontierSWEnot in index	–	–	–	25.3%#12	–	Epoch AI Benchmarking Hub
Terminal-Bench 4.0 (AA)not in index	3.0%#94	13.1%#65	21.2%#51	17.2%#57	–	Artificial Analysis
Terminal-Bench 2.1 (AA)not in index	75.3%#61	84.3%#30	88.4%#10	88.0%#12	–	Artificial Analysis

Agents & tools

Benchmark	low	medium	high	xhigh	not stated	Source
Terminal-Bench	–	–	20.3%#64	–	–	Terminal-Bench
APEX-Agents	–	–	–	–	65.3%#7	Mercor
LMArena Agent	–	–	–	1.3#24	–	LMArena
LiveBench Agentic Codingnot in index	–	–	–	–	57.0%#21	LiveBench
Terminal-Bench 2.1 (Vals)	–	–	78.3%#13	–	–	Vals AI
GDPval (AA)not in index	45.1%#64	55.2%#23	55.5%#22	56.6%#17	–	Artificial Analysis
τ²-Bench Banking (AA)	38.1%#38	44.3%#17	50.7%#2	43.3%#19	–	Artificial Analysis
Analyst Agent (AA)not in index	–	–	41.3%#16	–	–	Artificial Analysis
CyberBenchnot in index	–	–	67.7%#25	–	–	Vals AI
Finance Agent v2not in index	–	–	53.7%#26	–	–	Vals AI
Harvey's Legal Agent Benchmarknot in index	–	–	15.8%#6	–	–	Vals AI
Legal Research Benchnot in index	–	–	48.1%#8	–	–	Vals AI
SkillsBenchnot in index	–	–	55.8%#13	–	–	Vals AI
Tax Agent Benchnot in index	–	–	70.8%#8	–	–	Vals AI
Terminal-Bench 4.0 (Vals)not in index	–	–	17.2%#24	–	–	Vals AI
Terminal-Bench Sciencenot in index	–	–	11.4%#10	–	–	Vals AI
Time Horizon Index: KSPnot in index	–	–	5.8%#10	–	–	Vals AI
Vending-Bench 2not in index	–	–	–	–	9047.0#9	Andon Labs
AutomationBenchnot in index	–	–	66.7%#8	67.0%#7	–	Artificial Analysis
GDP.pdfnot in index	–	–	17.0%#33	17.2%#32	–	Artificial Analysis
EnterpriseOps-Gymnot in index	–	–	48.3%#5	–	–	Artificial Analysis
AA-Briefcasenot in index	–	–	1536#20	1544#17	–	Artificial Analysis

Maths

Benchmark	low	medium	high	xhigh	not stated	Source
FrontierMath Tiers 1–3	–	–	–	66.0%#28	–	Epoch AI Benchmarking Hub
FrontierMath Tier 4	–	–	–	31.7%#32	–	Epoch AI Benchmarking Hub
OTIS Mock AIME	–	–	97.8%#24	99.2%#14	–	Epoch AI Benchmarking Hub
ProofBench	–	–	–	–	51.0%#27	Vals AI
LiveBench Mathematicsnot in index	–	–	–	–	92.6%#25	LiveBench
LMArena Maths	–	–	1448#75	–	–	LMArena

Knowledge

Benchmark	low	medium	high	xhigh	not stated	Source
SimpleQA Verified	–	–	49.3%#27	48.9%#28	–	Epoch AI Benchmarking Hub
LiveBench Data Analysisnot in index	–	–	–	–	73.9%#38	LiveBench
AA-Omniscience	25.9#48	28#39	30.5#31	29.3#34	–	Artificial Analysis
MMLU-Pro	–	–	89.4%#11	–	–	Vals AI
LegalBench	–	–	86.3%#13	–	–	Vals AI
CorpFin	–	–	66.2%#20	–	–	Vals AI
TaxEval	–	–	71.1%#81	–	–	Vals AI
BioMysteryBenchnot in index	–	–	72.2%#8	–	–	Vals AI
Excel Modeling Benchmarknot in index	–	–	62.7%#28	–	–	Vals AI
MedCodenot in index	–	–	44.7%#38	–	–	Vals AI
MedScribenot in index	–	–	86.5%#20	–	–	Vals AI
Public Benefits Benchnot in index	–	–	66.8%#14	–	–	Vals AI
EBR-benchnot in index	–	–	–	30.5%#10	–	Epoch AI Benchmarking Hub
GDP.pdfnot in index	–	–	16.0%#34	17.2%#30	–	Epoch AI Benchmarking Hub
AA-Omniscience: accuracynot in index	43.3%#97	41.9%#104	48.2%#70	43.0%#100	–	Artificial Analysis
AA-Omniscience: non-hallucinationnot in index	69.3%#45	76.0%#17	65.7%#57	76.0%#18	–	Artificial Analysis

Instruction following

Benchmark	low	medium	high	xhigh	not stated	Source
LiveBench Languagenot in index	–	–	–	–	83.7%#18	LiveBench
LiveBench Instruction Followingnot in index	–	–	–	–	71.9%#20	LiveBench
LMArena Instruction Followingnot in index	–	–	1451#63	–	–	LMArena

Human preference

Benchmark	low	medium	high	xhigh	not stated	Source
LMArena Text	–	–	1454#71	–	–	LMArena
LMArena Creative Writingnot in index	–	–	1445#52	–	–	LMArena
LMArena Multi-turnnot in index	–	–	1444#97	–	–	LMArena
LMArena Longer Queriesnot in index	–	–	1471#59	–	–	LMArena

Multimodal

Benchmark	low	medium	high	xhigh	not stated	Source
LMArena Vision	–	–	1265#46	–	–	LMArena
MortgageTaxnot in index	–	–	64.2%#46	–	–	Vals AI
SAGEnot in index	–	–	28.9%#81	–	–	Vals AI
Blueprint-Bench 2not in index	–	–	–	–	33.2%#11	Epoch AI Benchmarking Hub

Long context

Benchmark	low	medium	high	xhigh	not stated	Source
AA-LCR	80.7%#58	81.0%#54	80.3%#63	81.0%#54	–	Artificial Analysis
LMArena Documentnot in index	–	–	1461#24	–	–	LMArena
MLCRnot in index	–	–	12.2%#28	–	–	Artificial Analysis

Composite

Benchmark	low	medium	high	xhigh	not stated	Source
Epoch Capabilities Indexnot in index	–	–	–	–	156.6#19	Epoch AI Benchmarking Hub
LiveBench	–	–	–	–	78.0%#17	LiveBench
AA Intelligence Index	35.1#77	42.8#43	44.3#37	44.2#39	–	Artificial Analysis
Vals Indexnot in index	–	–	52.1#20	–	–	Vals AI

Where to run it

Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.

Provider	Speed	First token	Input $/M	Output $/M	Context	Quantisation
Amazon Bedrock	96 tok/s	3.48 s	$2.20	$6.60	500k	–
SpaceXAI	58 tok/s	1.89 s	$2.20	$6.60	500k	–
SpaceXAI	56 tok/s	2.48 s	$2.00	$6.00	500k	–
SpaceXAI	54 tok/s	2.83 s	$2.00	$6.00	500k	–

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 at $0.500 per 1M. Reasoning tokens are not modelled.

Workload	Tokens in / out	Cost	With caching	Time
Chat reply	400 / 300	$0.0026	$0.0022	28.1 s
Summarise a 30-page report	12,000 / 600	$0.028	$0.014	32.0 s
Code edit	6,000 / 1,500	$0.021	$0.014	43.9 s
Agentic coding session	60,000 / 4,000	$0.144	$0.076	1.3 min
Structured extraction	2,000 / 200	$0.0052	$0.0030	26.7 s