Superpower Daily
Nvidia’s $92 Billion Q2 Forecast Puts Hyperscaler Dependence in...SemiAnalysis Says AgentX Drove 50-Plus Upstream Fixes for AI Ag...MiniMax H3 Opens Its Video Weights, but Not Where Many Develope...Together AI Secures $240 Million IBM Cloud Buildout for Open-So...Higgsfield Puts Its Video Engine Behind Grok Bot—and Keeps the...Anthropic Gives Enterprise Teams Mythos 5’s Bug Hunt, Not Its P...Nvidia’s $92 Billion Q2 Forecast Puts Hyperscaler Dependence in...SemiAnalysis Says AgentX Drove 50-Plus Upstream Fixes for AI Ag...MiniMax H3 Opens Its Video Weights, but Not Where Many Develope...Together AI Secures $240 Million IBM Cloud Buildout for Open-So...Higgsfield Puts Its Video Engine Behind Grok Bot—and Keeps the...Anthropic Gives Enterprise Teams Mythos 5’s Bug Hunt, Not Its P...
Productspublished

Vertex AI Adds Grok 4.6 Preview With a Lower Listed Cache Rate

Google Cloud gives buyers another route to Grok 4.6, but the model remains a Preview offering and the lower cached-input rate is specific to Vertex AI rather than a universal price cut.

By 2 min read
Vertex AI Adds Grok 4.6 Preview With a Lower Listed Cache Rate
Sign in to save and react

Listen to this story

The audio brief

About 1:30
0:001:30
Read transcript
Google Cloud has added Grok 4.6 to Vertex AI’s Model Garden, giving customers a new way to access the model—and a lower listed price for cached input. The Vertex AI rate is 30 cents per million cached tokens for prompts up to 200,000 tokens. SpaceXAI’s direct API lists 50 cents for the same tier. Standard input and output pricing is unchanged between the two routes: $2 per million input tokens and $6 per million output tokens. The distinction matters for applications that repeatedly send the same instructions or working context, but it is not a blanket discount on Grok 4.6. It applies to the Vertex AI listing, and actual savings depend on how much of a team’s traffic is cached. Vertex AI is now the fourth distribution route for the model, following SpaceXAI’s own API, GitHub Copilot, and Amazon Bedrock. There is also a threshold to model carefully. Above 200,000 tokens, SpaceXAI’s direct API lists $4 input, $1 cached input, and $12 output per million tokens. That matters because Grok 4.6 supports a 500,000-token context window. The main constraint is status: Google Cloud labels this Vertex AI offering Preview. So the announcement expands purchasing and deployment options, but does not establish general availability or production suitability. The thing to watch is whether that Preview route becomes a dependable production option for long-context, cache-heavy workloads.

Story brief

3 key points

Google Cloud’s Vertex AI Model Garden now offers SpaceXAI’s Grok 4.6, but only as a Preview distribution option. The listing keeps standard input and output rates at $2 and $6 per million tokens while lowering cached input to $0.30 per million, versus $0.50 on SpaceXAI’s direct API for prompts up to 200,000 tokens. The direct API’s pricing changes above that threshold. The opportunity is a potentially cheaper Google...

  1. 01

    Vertex AI’s August 21 addition creates a fourth route after SpaceXAI API, GitHub Copilot, and Amazon Bedrock launches.

  2. 02

    For prompts up to 200,000 tokens, Vertex lists $2 input, $6 output, and $0.30 cached input per million tokens.

  3. 03

    SpaceXAI’s direct API charges $0.50 per million cached tokens in that band, then lists $4/$1/$12 above 200,000.

Teams that want to buy Grok 4.6 through Google Cloud can now do so through Vertex AI Model Garden. The new route lists a lower cached-input price than SpaceXAI’s direct API for prompts up to 200,000 tokens, although Google Cloud currently classifies the offering as Preview.

A fourth distribution route

SpaceXAI announced Grok 4.6 availability through Vertex AI on August 21. Google Cloud’s release notes list it in Model Garden as a Preview offering, meaning the addition expands where customers can obtain the existing model rather than introducing a new model release.

The Google Cloud listing follows a quick widening of Grok 4.6 distribution. SpaceXAI released it through its own API on August 12; GitHub Copilot added it on August 14; and Amazon Bedrock made it generally available on August 19. Vertex AI gives organizations already using Google Cloud’s model platform another purchasing and deployment path.

The rate card changes at 200,000 tokens

The direct API’s lower schedule has a boundary. For prompts above 200,000 tokens, SpaceXAI lists direct-API rates of $4 per million input tokens, $1 per million cached input tokens, and $12 per million output tokens. That threshold is relevant because SpaceXAI says Grok 4.6 supports a 500,000-token context window—the amount of text a model can consider in one request.

Cached input generally matters when an application sends the same material repeatedly, such as persistent instructions or accumulated working context. The listed price alone does not determine a workload’s bill: actual cost depends on prompt length and on how often requests use cached tokens. Buyers comparing the two routes therefore need to model their own request patterns rather than treating the 20-cent difference as a blanket discount.

Preview status remains the practical constraint

For now, the Vertex AI route is explicitly a Preview offering. The immediate decision is less about whether Grok 4.6 exists on Google Cloud than whether a Preview service fits a team’s production requirements. The announcement establishes availability and a listed rate card; it does not turn the Google Cloud route into a generally available service.

Related coverage

Continue reading