Grok 4.5 Shows Exceptional Vulnerability Detection – A New Tool for Security Researchers

1. Release Overview

xAI announced on July 8, 2026 that Grok 4.5 is its most advanced model to date, designed for coding, agent tasks, and knowledge work, and trained jointly with the AI programming editor Cursor.

The model was trained on tens of thousands of NVIDIA GB300 GPUs using large‑scale reinforcement learning across hundreds of thousands of tasks, most of which focus on multi‑step software‑engineering and other technical work.

2. Benchmark Performance

DeepSWE 1.0 (pass@1): Grok 4.5 scores 62.0%, versus Fable 66.1%.

DeepSWE 1.1: Grok 4.5 scores 53%, versus Fable 70%.

Terminal Bench 2.1: Grok 4.5 scores 83.3%, versus Fable 84.3%.

SWE Bench Pro (solution rate): Grok 4.5 scores 64.7%, versus Fable 80.4%.

Grok 4.5 also demonstrates a striking token‑efficiency advantage: on SWE Bench Pro it uses an average of 15,954 output tokens, while Opus 4.8 (max) consumes 67,020 tokens—a 4.2× difference.

3. Vulnerability Detection Validation

Security researcher @Zaddyzaddy posted on social media that "Grok 4.5 exhibits an unbelievably strong capability in vulnerability detection."

The model is now integrated into all Cursor subscription tiers and set as the default model in Grok Build. Its API pricing is $2 per million input tokens and $6 per million output tokens, with an inference speed of 80 TPS.

4. Conclusion

The launch of Grok 4.5 marks a new stage for AI applications in software engineering and security research, offering powerful vulnerability‑detection ability and high token efficiency as a fresh automated tool for security researchers.