Tested Claude 4 Opus vs Grok 4 on 15 Rust coding tasks - actual numbers inside

Ran both models through identical coding challenges on a 30k-line Rust codebase. Here's what the data shows:

**Bug Detection:** Cursor 4 caught every race condition and deadlock I threw at it. Opus missed several, including a tokio::RwLock deadlock and a thread drop that prevented panic hooks from executing.

**Speed:** Cursor averaged 9-15 seconds, Opus 13-24 seconds per request.

**Cost:** $4.50 vs $13 per task. But Cursor's pricing doubles after 128k tokens.

**Rate Limits:** Cursor's limits are brutal. Constantly hit walls during testing. Opus has no such issues.

**Tool Calling:** Both at 99% accuracy with JSON schemas. XML dropped to 83% (Opus) and 78% (Cursor).

**Rule Following:** Opus followed my custom coding rules perfectly. Cursor ignored them in 2/15 tasks.

**Single-prompt success:** 9/15 for Cursor, 8/15 for Opus.

**Bottom line:** Cursor is faster, cheaper, and better at finding hard bugs. But the rate limits are infuriating and it occasionally ignores instructions. Opus is slower and pricier but predictable and reliable.

For bug hunting on a budget: Cursor. For production workflows where reliability matters: Opus.

Full breakdown [here](https://forgecode.dev/blog/cursor-4-opus-vs-cursor-4-comparison-full/)

Anyone else tested these on real codebases? Curious about experiences with other languages.