xAI released Grok 4.5, trained on tens of thousands of Nvidia GB300 GPUs. The model trails Fable 5 and GPT-5.5 on coding benchmarks but offers a radical price advantage that reshapes the cost-performance equation for enterprise AI.
Grok 4.5 costs $2 per million input tokens. This undercuts competitors significantly. The model requires 4.2 times fewer tokens than Claude Opus 4.8 to solve comparable tasks, meaning actual per-task costs drop further. For developers and teams running inference at scale, this efficiency matters more than raw benchmark scores.
Benchmark performance matters less when the cost gap widens this far. Fable 5 and GPT-5.5 may score higher on standardized coding tests, but they cost substantially more per token and per task. Organizations building AI products with tight margins face a different calculus. A model that costs one-tenth as much per inference becomes viable for use cases previously impossible to monetize.
xAI trained Grok 4.5 on a cluster exceeding 100,000 Nvidia GB300 GPUs. The GB300 represents Nvidia's latest inference accelerator, designed for production workloads at hyperscale. This infrastructure investment signals xAI's commitment to competitive performance within its cost structure.
EU availability launches mid-July, expanding access beyond North America. Regional availability drives adoption among European enterprises bound by data residency rules.
The release reflects a broader competitive shift in AI markets. Pricing pressure from rivals has forced every major lab to optimize for cost-efficiency, not just capability. Grok 4.5 weaponizes this approach. It trades marginal capability gains for dramatic price reductions, betting that thousands of teams need good-enough models at 80 percent lower