Grok 4.5
Points and comments are a snapshot, not live.
SpaceXAI launched Grok 4.5, its strongest model for coding and agentic tasks.
Grok 4.5 was trained on tens of thousands of NVIDIA GB300 GPUs with extensive data filtering and RL on multi-step engineering tasks. It achieves 80 TPS and 4.2x token efficiency over Opus 4.8 on SWE Bench Pro. Pricing is $2/M input and $6/M output tokens (under 200K context; above that doubles). The model is available in Grok Build, Cursor, and the SpaceXAI API, but not yet in the EU. Benchmarks show competitive scores on DeepSWE, Terminal Bench, and SWE Bench Pro.
What commenters are saying
Commenters debated Grok's popularity and political baggage, with many citing supply chain risk from Elon Musk's platform weaponization. Several users praised Grok Build's CLI as fast and responsive, but noted previous models were outclassed at SWE until now. Token economics drew attention: at $2/$6 with higher efficiency, Grok 4.5 may undercut competitors. However, the doubled pricing above 200K context was flagged as notable and omitted from the announcement.