xAI's latest model, Grok 4.5, has achieved a remarkable 29.0% resolution rate on the SWE Marathon benchmark, showcasing its dominance in AI-driven software engineering tasks since its launch on July 8, 2026.
This impressive figure positions Grok 4.5 ahead of competitors such as Claude Opus 4.8, which scored 26.0%, and Fable at 24.0%. Additionally, Grok 4.5 recorded an 83.3% resolution rate on the Terminal-Bench 2.1, focusing on terminal-based development workflows.
As for pricing, Grok 4.5 is offered at $2 per million input tokens and $6 per million output tokens via its API, claiming about 2x better token efficiency than leading rivals. This means that for tasks requiring 200,000 tokens by competitors, Grok 4.5 may complete the same task in around 100,000 tokens.
Just a week after Grok 4.5's launch, Moonshot AI introduced Kimi K3, which features 2.8 trillion parameters. While Kimi K3 is positioned as a competitor with its open-weight system, allowing developers to modify model weights, it is noted to be slower but more cost-effective.
Both models lack blockchain integration or token utility, suggesting limited direct impact on cryptocurrency markets. However, Kimi K3’s architecture may facilitate broader experimentation and increased compute demand as developers create custom versions.
This article is for informational purposes only and is not financial advice.
