NASA’s new dark energy space telescope can also detect killer asteroids
technologyreview.com·2h ago
Cloud & Infrastructure
The New Stack·17d ago·5 min read
I trust Claude for everything. This test made me rethink that.
TL;DR
A recent test comparing Grok 4.5 and Opus revealed surprising results regarding token efficiency. Grok 4.5, released by xAI, claims to match Opus's performance using only a quarter of the tokens.
✦ Why It Matters
Evaluate Grok 4.5 for your next project to potentially reduce token usage and costs.
Key Takeaways
- Test Grok 4.5 for coding tasks to reduce costs.
- Evaluate token usage when choosing AI models.
- Consider Grok's performance in real-world coding scenarios.
- Monitor pricing changes for AI models to optimize expenses.
Full Summary
xAI's Grok 4.5 was released with the claim that it can perform coding tasks as well as Claude Opus 4.8 but at a fraction of the token usage and cost. In a series of tests involving bug fixes, refactoring, and feature builds in a Rust codebase, Grok consistently produced identical results to Opus while using 4.3 times fewer tokens and costing about 80% less.
For instance, in a bug fix task, Grok took 46.25 seconds and 210.2K tokens, while Opus completed it in 30.43 seconds but used 174.1K tokens. Overall, Grok completed all tasks in approximately 3 minutes and 46 seconds, costing $1.00, compared to Opus's 11 minutes and 40 seconds at $5.14.
These results suggest that Grok is a viable alternative for developers looking to optimize their token usage without sacrificing quality.