Grok 4.5 lands in Cursor, with strong early praise and usage concerns

Cursor now offers Grok 4.5. Cursor says it trained Grok 4.5 with SpaceXAI, calls it its strongest model so far, and presents it as the first model it built for more than software engineering. The listed price is $2 per million input tokens and $6 per million output tokens, with double usage during the first week after launch.

Early experience is mixed. Grok 4.5 High Fast has been used for product requirement documents and issue writing with results described as close to Opus, much faster, and less padded than Opus or GPT5.5. At the same time, some users on the latest Cursor version cannot see Grok 4.5, and some saw it appear on launch day and then disappear later.

Availability may also differ by region, with an EU user specifically reporting that the model was missing. Other reports raise billing and routing concerns: Grok 4.5 appeared to use GPT 5.6 sol as a subagent in one case, while Grok 4.5 High Fast reportedly switched to Sonnet 5 High and consumed on-demand usage in another. The open practical questions are how manual model choice compares with Auto mode, how Grok 4.5 counts against Cursor’s $20 plan, and how that mileage compares with Codex’s $20 plan for daily coding.