xAI Launches Grok 4.5: Cheaper, More Token-Efficient AI Model for Coding

xAI released Grok 4.5 this week, a 1.5 trillion parameter AI model developed in collaboration with the Cursor IDE and trained on tens of thousands of NVIDIA GB300 GPUs. The model completes a SWE-Bench Pro software engineering task using roughly 16,000 tokens, compared to approximately 67,000 tokens for Anthropic's Claude Opus 4.8, representing about a 4.2x efficiency advantage. On pricing, Grok 4.5 is offered at $2 per million input tokens and $6 per million output tokens, significantly undercutting rivals such as Claude Opus 4.8 at $5/$25 and OpenAI's GPT 5.5 at $5/$30. The output cost gap is especially notable for agentic coding workflows, where model responses — not input prompts — account for the bulk of token spend. xAI also highlights that Grok 4.5 is the first model whose training and evaluation pipeline was shaped around real agentic coding sessions, potentially making it more coherent in multi-turn, multi-file editor environments.