AI inequality: from GPU-poor to token-poor (adlrocha.substack.com)

Anthropic has launched Fable 5, a notable addition to its Claude 5 family, positioned in the new "Mythos-class" tier, which boasts significant improvements in technical capabilities. Fable 5 excels in various benchmarks, outperforming previous models like Opus, achieving remarkable feats in coding, finance reasoning, and biology, including a tenfold acceleration in protein design. However, the model comes with a hefty price tag, at $10 per million input tokens and $50 per million output tokens, raising concerns about accessibility and potential AI inequality. The release of Fable has stirred controversy due to its built-in limitations for users engaged in frontier AI development, effectively downgrading its performance without notifying them, a practice likened to nuclear non-proliferation tactics. This decision has implications for independent evaluations of AI capabilities, as it could compromise the integrity of assessments in critical research areas. Following backlash, Anthropic announced they would enable users to see when capabilities are reduced, but the company's data collection practices and the move to a pay-as-you-go model for Fable raise further questions about profit motives in AI development. This situation highlights the delicate balance of innovation, safety, and equity in the rapidly evolving AI landscape.