API model
grok-4.6
Use the exact ID; aliases and partner labels are separate.
SpaceXAI model record · checked August 20, 2026
Grok 4.6 is the current model; access is surface-specific.
Use grok-4.6 for the direct SpaceXAI API. Evaluate its 500K context, reasoning controls, tools, caching, and completed-task cost on real work—without extending provider availability to Bedrock, consumer plans, partners, or Flowith.
Official API facts
Context
500,000 tokens
Provider-documented window, not a reason to skip retrieval and compaction.
Reasoning
low · medium · high · xhigh
High is the direct API default documented at source check.
Modalities
text + image → text
Do not infer image generation or every consumer-app feature.
Choose by workload contract
Is the job long-running coding or knowledge work?
Pilot Grok 4.6 on a representative task with tests and a rollback model.
Does the prompt repeat across many turns?
Set a stable prompt cache key and verify actual cache hits and billed tokens.
Will tools produce a long trajectory?
Budget tool calls, compact context deliberately, and retain an inspectable trace.
Must inference stay on AWS?
Use the Bedrock route; endpoint, model ID, features, region, IAM, and price differ.
Grok 4.5 remains a useful version-specific baseline. Compare task success, tests, latency, input and output tokens, cache hits, tool calls, review effort, and rollback behavior before migration. For AWS-hosted inference, continue to the Bedrock endpoint and regions guide or run the deployment readiness checker.
Primary sources and boundaries
Grok 4.6 developer guide
Model ID, context, modalities, reasoning, APIs, tools, direct price, caching, and provider surfaces.
Grok 4.6 announcement
Release record, provider positioning, demos, reported evaluations, and named launch surfaces.
Provider benchmark results are provider claims, not independent guarantees. Recheck model lifecycle, live price, account rate limits, tools, regions, retention, and signed-in entitlement before production use.
Grok 4.6 questions
Grok 4.6 is SpaceXAI's model released August 12, 2026 for coding, agentic tasks, and knowledge work. The provider documents grok-4.6 as its current code and general-purpose API model.
The direct SpaceXAI API model ID is grok-4.6. Amazon Bedrock also uses xai.grok-4.6 plus endpoint-specific inference IDs; those are separate contracts.
At the August 20 source check, SpaceXAI listed $2 per million input tokens and $6 per million output tokens for the direct API. Recheck live pricing, cache behavior, tools, and partner markups before budgeting.
Yes, the official developer guide lists 500,000 tokens. Effective task quality and cost still depend on retrieval, prompt ordering, cache hits, tool output, compaction, and the requested output.
This page does not claim a Flowith launch or entitlement. Direct API, Grok Build, Cursor, gateways, Amazon Bedrock, consumer products, and Flowith are separate surfaces; verify the live workspace.