Summary
On July 8, 2026, xAI released Grok 4.5, its first model built specifically for coding and agentic work, co-trained on real Cursor developer sessions. Priced at $2/$6 per million tokens, it lands fourth on the Artificial Analysis Intelligence Index while costing over 60% less than Claude Opus 4.8 or GPT-5.5.
What changed
xAI shipped Grok 4.5 (model id grok-4.5), built on its 1.5-trillion-parameter V9 foundation and trained on real Cursor session data across hundreds of thousands of multi-step software-engineering tasks. It is available in the xAI API, xAI's Grok Build coding harness, and Cursor on all plans, with third-party access via OpenRouter, Vercel, Cloudflare, Snowflake, and Databricks Mosaic. Pricing is $2 input and $6 output per million tokens.
Why it matters
A capable coding and agentic model at roughly a third of Opus/GPT pricing pressures the cost of running long agent sessions, where token spend dominates. Co-training with Cursor also shows model builders optimizing directly against a specific coding-agent harness, tightening the loop between model and tool.
Evidence excerpt
It is an Opus-class model, but faster, more token-efficient and lower cost. ... trained on real Cursor session data for coding and agentic tasks ... available in the xAI API under the model name grok-4.5 ... and in Cursor across all plans.