Google's August 21 release notes list the model as a Preview offering, while SpaceXAI quotes $0.30 cached input and a 500,000-token context window.
By RuntimeWire Staff · Published
Primary source: SpaceXAI
Why it matters #
Google's August 21 release notes list Grok 4.6 as a Preview offering in Model Garden, extending the model's distribution beyond xAI's API, GitHub Copilot and Amazon Bedrock. The $0.30 cached-input rate advertised by SpaceXAI for Vertex AI is lower than the direct API's $0.50 rate for prompts up to 200,000 tokens.
Elon Musk-founded xAI, now branded SpaceXAI, announced on August 21 that Grok 4.6 was available through Google Cloud's Vertex AI. Separately, Google's August 21 release notes list xAI's Grok 4.6 as available in Preview in Model Garden.
The August 21 announcement is a distribution update for an existing model: xAI released Grok 4.6 on August 12, then added GitHub Copilot, Amazon Bedrock and Vertex AI availability. SpaceXAI directs developers to a Model Garden listing and a Google-hosted model card.
The $0.30 rate comes from SpaceXAI
Under its public-facing SpaceXAI branding, xAI advertises cached input at $0.30 per million tokens through Vertex AI. Its August 21 announcement also lists $2 per million input tokens and $6 per million output tokens.
The xAI API release notes set a different schedule for direct access. For prompts of up to 200,000 tokens, Grok 4.6 costs $2 per million input tokens, $0.50 per million cached input tokens and $6 per million output tokens. Above that threshold, the rates rise to $4, $1 and $12, respectively.
The $0.30 and $0.50 cached-input figures should not be treated as a single platform-wide rate. The lower figure applies to the Google route described by SpaceXAI, while the higher figure applies to xAI's direct API. Google's release-note entry establishes Preview availability without listing pricing.
That difference can matter in long-running agent workloads, where repeated system instructions, repository files and accumulated tool results can make cached tokens a meaningful share of the bill. Actual spending depends on prompt length and cache-hit rates.
Grok 4.6 carries a 500,000-token context window
SpaceXAI says Grok 4.6 has a 500,000-token context window and four configurable reasoning settings: low, medium, high and xhigh. Its August 12 launch post says the model received a longer supplemental training run than Grok 4.5, followed by supervised fine-tuning and reinforcement learning across coding, knowledge work, science and other technical tasks.
The same announcement reports a 69.9% score on CursorBench 3.2, compared with 66.7% for Grok 4.5 and 67.2% for GPT-5.6 Sol. Those are SpaceXAI-published benchmark results, and the company says third-party scores use the best self-reported or publicly available figures. The material supplies no comparative latency, reliability, safety, cost or long-context performance data.
Anthropic positions Claude Opus 4.5 for coding, computer use and long-running enterprise tasks, while OpenAI describes GPT-5 as a model for coding and agentic work. The available material does not support a head-to-head judgment among those models and Grok 4.6.
xAI has the compute; enterprise adoption is still the test
Musk helped establish xAI in July 2023 with researchers whose prior affiliations included DeepMind, OpenAI, Google Research, Microsoft Research, Tesla and the University of Toronto. The founding announcement connected members of the team to earlier work on systems including AlphaStar, AlphaCode, GPT-3.5 and GPT-4. xAI's careers site places its principal base in Palo Alto and lists hiring operations in San Francisco, Seattle, Austin, Memphis and London.
xAI announced a $20 billion Series E on January 6, exceeding its stated $15 billion target. Named investors included Valor Equity Partners, StepStone Group, Fidelity Management & Research, Qatar Investment Authority, MGX and Baron Capital Group. NVIDIA and Cisco Investments participated as strategic investors. xAI did not identify a lead investor.
xAI said the financing would support compute infrastructure and reported more than 1 million H100-equivalent GPUs across its Colossus I and II systems. It also claimed approximately 600 million monthly active users across X and the Grok apps at the end of 2025 and said Grok 5 was in training. The combined audience figure does not disclose Grok's standalone usage or adoption of Grok 4.6.
Grok 4.6 reached GitHub Copilot on August 14 and became generally available through Amazon Bedrock on August 19. SpaceXAI also made it available through its own API on August 12. Google's dated release note now lists the model as a Preview offering in Model Garden.