Grok 4.5: Cheap AI That Holds Its Own

xAI just dropped Grok 4.5, and its pricing is aggressively low: $2 per million input tokens versus $10-$50 for rivals like Fable 5 and GPT-5.5. Benchmark performance is mixed but respectable—on Terminal Bench 2.1, it nearly ties the leaders, scoring 83.3% vs 83.4% for GPT-5.5 and 84.3% for Fable 5. However, on tougher software engineering tests like DeepSWE 1.1, it lags at 53%, well behind Fable 5’s 70%. xAI focused on quality data filtering and asynchronous reinforcement learning to keep training efficient. The model also uses fewer tokens per task, making it even cheaper overall. This pricing strategy mirrors what Chinese firms have done: offer solid performance at a fraction of the cost. Why it matters: Grok 4.5 signals that cost could become the deciding factor in AI adoption, potentially disrupting the premium pricing models of OpenAI and Anthropic.