On intelligence per dollar basis Grok 4.5 looking brilliant.
AI. At $0.31 per Intelligence Index task (of Artificial Analysis), it delivers near-frontier performance at roughly 1/9 the cost of Claude Fable 5, 1/6 the cost of Claude Opus 4.8, and about 1/3 the cost of GPT-5.6 Sol or Kimi K3.
That's more relevant with the token-consumption data. Grok 4.5 uses nearly 6X fewer tokens than Fable 5 while still being near the top of the ranking, onFrontierSWE.
For many real-life AI applications this is better than a small lead on a benchmark.
Lower cost per completed task means cheaper agents, higher usage limits, faster testing and better margins at scale.
The frontier model race must shift from "raw intelligence" to "efficient intelligence".
Grok 4.5 is arguably #1 when taking speed and cost into account