Grok 4.7 Release Nears as xAI Adds SpaceX Training Data

SAN FRANCISCO: The Grok 4.7 release could arrive in early September, after Elon Musk said initial training was complete and that supplemental training was adding a large amount of SpaceX data. Musk said on August 12 that the model should be ready in '3 to 4 weeks,' putting the indicated window at roughly September 2-9. However, xAI had not published an official Grok 4.7 launch page or release note as of September 1. Musk has described Grok 4.7 as significantly better than Grok 4.6. Earlier founder statements also described the upcoming model as having about 2.1 trillion parameters and being slightly slower to serve while improving token efficiency. xAI has not yet published a model card confirming those specifications. Interesting. Grok 4.6 releases around August 7. This will be the 1.5T model with significantly improved SFT & RL. Grok 4.7 will be the 2.1T model released a few weeks later. This will be better than 4.6 in every way, except slightly slower to serve, albeit with even better… https://t.co/dJjXVnFwoA — Elon Musk (@elonmusk) July 28, 2026 Claims that the additional training specifically includes Starlink telemetry, rocket test logs or particular engineering records remain less firmly established. Musk's publicly reported statement referred more broadly to a 'massive amount of SpaceX company data.' Read: Grok Deepfake Scandal: Why Elon Musk's xAI Is Facing Global Backlash The current Grok 4.6 model launched on August 12 for coding, agentic tasks and knowledge work. xAI lists a 500,000-token context window and API pricing of USD 2 per million input tokens and USD 6 per million output tokens. Artificial Analysis scored Grok 4.6 at 61 on its Intelligence Index, level with GPT-5.6 Sol Max in that evaluation. Its measured output speed was about 59 tokens per second for the high-reasoning configuration. Read: Imagine Image 2.0 Adds Precise Editing Across Grok Apps No independent benchmark results for Grok 4.7 were available before launch. Claims that it will become the leading coding model therefore remain predictions rather than measured results.