Expired60% confidenceFactExact time
Because an early snapshot of the Cursor codebase was inadvertently included in the training data, the team did not use Cursor Bench to publicly promote Grok 4.5.
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
7/10/2026
First Seen
Valid until: 10/8/2026(expired)
Sources
Grok 4.5 Deep Dive: Coding on Par with Top Models, at a Fraction of the Cost
youtubeTheo - t3․gg7/9/2026
Related Claims
UnverifiedThird-party monitoring account LuminaBench reports that Grok 4.8 has appeared in Cursor's server-side model list, but neither xAI nor Cursor has officially confirmed this.68% similarUnverifiedThe only official information Cursor has made public is a mention at a developer conference in June that a model was being trained from scratch.67% similarUnverifiedCursor has launched CursorBench 4.0 for evaluating agent performance on real-world, ambiguous, multi-file tasks, and the addition of long-horizon problems has caused scores to drop across all models.66% similarUnverifiedCursor does not train foundation models itself; its technical architecture relies on calls to external large language model APIs.66% similarUnverifiedCursor maintains a private evaluation set called cursor bench, which is based on real engineering tasks from its own codebase and is strictly isolated from training data.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/427594API
curl https://kongchang.com/api/v1/knowledge/claims/427594MCP
get_claim(id=427594)