SpaceXAI has released Grok 4.6, a vision-language model developed in collaboration with Cursor and optimized for long-running agentic tasks. The model is available via API, Grok Build, and Cursor, featuring adjustable reasoning levels and support for up to 500,000 input tokens.
- Architecture: Approximately 1.5 trillion parameters.
- Training: Pretrained on curated data including anonymized coding-agent data from Cursor, followed by fine-tuning using transcripts generated by Grok 4.5 and reinforcement learning on agentic tasks.
- Performance: Tied for third on Artificial Analysis’ Intelligence Index (61) and achieved a top score of 94.9 percent on GPQA Diamond.
- Pricing: $2.00/$0.50/$6.00 per million input/cached/output tokens, with a fast mode available at higher rates.
The model aims to provide high capability for coding and knowledge work at a lower cost than some competing models, leveraging synthetic data and reinforcement learning to improve performance on complex tasks.