Grok 4.5 Is Here: Benchmarks, Pricing, 500K Context, and Real CLI Demos
The video reviews the newly released Grok 4.5 model from SpaceX AI, covering the announcement, benchmarks, pricing, and hands-on demos, including its first training collaboration with Cursor. It highlights benchmark positioning (e.g., between GPD 5.5 and Opus 4.8 on select tests), strong token efficiency on Swebench Pro, and a notable cost advantage, citing Artificial Analysis metrics like ~$0.31 cost per task, ~91 output tokens speed, and #4 on the intelligence index. Training details include tens of thousands of NVIDIA GB300s, heavy data filtering/curation, and reinforcement learning over hundreds of thousands of multi-step software engineering tasks. The model offers a 500K context window, API pricing of $2/M input and $6/M output, limited-time free access via Grok Build CLI and some Cursor areas, and a demo generating a SaaS landing page plus a /goal command example consuming 2.5M free tokens.
00:00 Grok 4.5 Overview
00:21 Benchmarks Cost Speed
01:19 Training Details
01:53 Blog Demos Token Efficiency
02:58 Pricing And Access
03:35 More Benchmarks Future Models
05:51 Context Window And CLI
06:39 Landing Page Demo
08:41 Free Tokens Agent Mode
09:31 Wrap Up And Thanks
source
