SpaceXAI Launches Grok 4.5: The Price-Performance Gambit for AI Coding

Updated

SpaceXAI Launches Grok 4.5: The Price-Performance Gambit for AI Coding

On July 8, 2026, Elon Musk's SpaceXAI (the company formerly known as xAI) officially launched Grok 4.5, its latest flagship mixture-of-experts (MoE) reasoning model optimized for coding, autonomous agentic workflows, and professional knowledge work. While the model delivers highly competitive price-to-performance metrics, its launch has been severely overshadowed by a massive data privacy scandal involving its "Grok Build" development CLI, prompting an unprecedented data purge and a sudden pivot to open-sourcing the tool.

Model Architecture and Benchmarks

Grok 4.5 is a 1.5-trillion-parameter MoE model featuring a 500,000-token context window. It was jointly trained with coding platform Cursor on trillions of tokens of developer and agent interactions.

  • API Pricing: Billed at $2.00 per million input tokens and $6.00 per million output tokens (for the first 200,000 context tokens), with cached inputs priced at $0.50 per million.
  • Benchmarks: Grok 4.5 leads the SWE Marathon benchmark at 29.0%, beating Anthropic's Claude Opus 4.8 (26%) and Fable. It also ranked first on the Harvey Legal Agent Benchmark and scored a 29% mean pass rate on Snorkel AI's professional GDPval+ evaluation (outperforming GPT-5.5 at 22%).
  • Speed: Independent evaluations by Artificial Analysis measured output speeds of 120 tokens per second, though with a notable 14.5-second time-to-first-token.

The Grok Build Privacy Scandal and Data Purge

In mid-July 2026, a security researcher discovered that the "Grok Build" command-line coding assistant was silently uploading entire local Git repositories—including full commit histories, proprietary source code, API keys, database passwords, and cloud credentials—to an xAI-controlled Google Cloud Storage bucket.

  • Data Overreach: In one documented test, Grok Build uploaded 5.1 gigabytes of data to complete a coding task that required only 192 kilobytes—approximately 26,000 times more data than necessary.
  • The Purge: Following public outcry, SpaceXAI CEO Elon Musk announced an absolute deletion of all historical user data on July 13, 2026:

    "As a precautionary measure, all user data that was uploaded to SpaceXAI before now will be completely and utterly deleted. Zero anything whatsoever will remain." — Elon Musk on X

  • Open-Sourcing to Rebuild Trust: To restore developer confidence and provide full operational transparency, SpaceXAI open-sourced the "Grok Build" coding agent harness and terminal user interface (TUI) on GitHub on July 15, 2026.

Legal Precedent: Suing Users Over Harmful Content

In a rare legal move, on July 15, 2026, SpaceXAI filed a federal lawsuit against a South Carolina user (Terry Harwood) for allegedly using Grok to bypass safety filters and generate explicit child sexual abuse material (CSAM). This marks one of the first major legal actions where an AI developer has sued its own customer for exploiting its platform to generate harmful content, signaling a aggressive shift in how AI firms enforce terms of service and manage liability.

Part of

This finding is an example of a pattern recurring across your work:

Revision history

  • Update the Grok model releases note with detailed specs on Grok 4.5, benchmarks, the massive Grok Build privacy scandal, Elon Musk's database purge, the open-sourcing of Grok Build on GitHub, and the landmark CSAM lawsuit.
    · by the agent
  • Updating the Grok 4.5 release finding with concrete pricing, benchmark data, and the strategic implications of the Cursor acquisition.
    · by the agent
  • Refine and consolidate the Grok 4.5 launch finding with the latest details on the Cursor codebase training leak, Tesla's mandated switch memo, and the $200 weekly cap controversy.
    · by the agent
  • Refine and consolidate the Grok 4.5 launch finding with the latest details on the Cursor codebase training leak, Tesla's mandated switch memo, and the $200 weekly cap controversy.
    · by the agent
  • Refine and consolidate the Grok 4.5 launch finding with the latest details on the Cursor codebase training leak, Tesla's mandated switch memo, and the $200 weekly cap controversy.
    · by the agent
  • Document the official launch of Grok 4.5, its pricing, benchmarks, and vertical integration with Cursor.
    · by the agent
  • Document the official launch of Grok 4.5, its pricing, benchmarks, and vertical integration with Cursor.
    · by the agent
  • Document the official launch of Grok 4.5, its pricing, benchmarks, and vertical integration with Cursor.
    · by the agent
  • Create new finding tracking xAI's Grok 4.5 private beta, Cursor integration, and Musk's monthly model release commitment.
    · by the agent