Skip to main content
TokenCost logoTokenCost

Grok 4.6

Last verified August 30, 2026 · xAI pricing

$2.00/1M input · $6.00/1M output · 500K context · xAI

Count Tokens

Tokens
Input Cost
Output Cost

Estimate Monthly Cost

Monthly Cost Estimator

Quick:
<$0.0001/mo

Pick a preset above or enter custom usage

Alternatives to Grok 4.6

Pricing Details

xAINow sold in four places at one headline card. Re-checked 2026-08-30 after the three-cloud rollout: Bedrock GA August 18, Google Model Garden August 21, Microsoft Foundry public preview August 26. All four venues list $2.00 input / $0.50 cached / $6.00 output, and the entire spread between them is the 200K tier. xAI and Google Vertex both double every line above the threshold to $4.00/$1.00/$12.00. AWS Bedrock's model card publishes NO context tier at all: Global CRIS $2.00 input / $6.00 output / $0.50 cache read, and In-Region and Geo CRIS $2.20 / $6.60 / $0.55, both flat across the full 500K window, with Priority at 1.75x and Flex at 0.5x. So Bedrock Global is exactly 0.5000x of xAI's long-context card on all three lines, and Bedrock's Priority tier is 0.875x of xAI's standard long-context rate. Azure Foundry also publishes a flat card ($2.00/$6.00/$0.50, Global Standard only, and only in a Foundry blog post - the official Azure pricing page has no Grok 4.6 row), but Microsoft's models-sold-directly-by-azure doc gives grok-4.6 a 200,000-token context window and 128,000 max output, not 500,000. Azure's ceiling sits exactly at the threshold, so the surcharge is unreachable there and Azure sells 40% of the window at the full headline price. Boundary wording differs: xAI says 'at 200k tokens and above', Google says 'longer than 200K', a one-token disagreement at exactly 200,000. OpenRouter applies the 200K doubling to its Bedrock endpoint too, which contradicts AWS's own card; likely a model-level override inherited onto the reseller row. OpenRouter also reports max_completion_tokens 450,000 for this model against the 16384 carried below. xAI is now branded SpaceXAI on docs.x.ai and OpenRouter while the x.ai domain and x-ai/ prefix are unchanged. Launched August 12, 2026 on Grok 4.5's exact rate card except for one line. Input $2.00 and output $6.00 are identical to Grok 4.5, as is the 500K context and the 200K threshold; cached input is $0.50/1M against Grok 4.5's $0.30, a 67% increase that takes the cache discount from 85% off to 75% off. Above 200K prompt tokens every rate doubles to $4.00/$12.00 (cached $1.00) and xAI bills the higher rate across all tokens in the request, not just the overage. Cached tokens count toward the 200K trigger, so a long cached prefix can cross the cliff on its own. Batch: none, and not merely 0% - xAI's batch docs state grok-4.6 and grok-4.5 requests are rejected. Priority Processing is a flat 2x on all token types applied after the cache discount, with no separate slug; xAI's announcement mentions a "fast variant" at twice the price without naming it, and Priority Processing is the only 2x option in the docs. Server-side tools: web/X search $5 per 1,000 calls. New xhigh reasoning effort. Knowledge cutoff February 1, 2026. No published aliases; grok-build-latest still resolves to Grok 4.5. No cache TTL published. maxOutput 16384 is not a provider figure: xAI states "no text output limit" for this model, so the number here is carried for schema consistency with the rest of the Grok line. Artificial Analysis Intelligence Index 60.92 (#6), Terminal-Bench 2.1 88.39, cost per index task $0.8367 against Grok 4.5's $0.3601. Terminal-Bench 3.0 (official Harbor board, read 2026-08-17): 26.5% +/- 1.5 on the Grok Build harness, rank 4 of 10, $2,095.44 for the 370-trial run, which is $21.37 per solved task and the cheapest on that board among rows whose cost figure the authors do not flag. Do NOT put the 88.39 and the 26.5 in the same table: 2.1 is saturated and 3.0 is a different, harder task set. Estimated tokens (different tokenizer).
Input / 1M tokens
$2
Output / 1M tokens
$6
Context Window
500K
Max Output
16K

Price History

No price change recorded since 2026-08-15, when this site started tracking this model; it launched on 2026-08-12.

Frequently Asked Questions

Common questions about Grok 4.6 pricing and usage

Read More About Grok 4.6