Anthropic Claude Fable 5 benchmark and price: 2x cost, 5.7% gain

Anthropic's Claude Fable 5 topped the Artificial Analysis Intelligence Index this week, posting a score of 64.9 and setting records in five of the index's ten component benchmarks. The model's lead over Stability AI's Opus 4.8 is modest in performance terms, about 5.7 percent, while its price is roughly double Opus 4.8 on comparable inference tiers.

Claude Fable 5's AAII result is the clearest metric available: 64.9 points on a composite designed to weigh reasoning, knowledge, coding, and safety metrics. The Fable 5 result beat the nearest competitor by the stated margin and produced top marks in half the tested categories, giving Anthropic a headline win on the index even as the absolute margin remained small.

Performance and benchmarks

Claude Fable 5 produced record scores in five of ten AAII subbenchmarks, according to the published index numbers. Those records are concentrated in higher-level reasoning tasks, where the model's output aligned better with benchmark gold standards, and in a subset of coding tasks measured by the index. The composite AAII score of 64.9 translates to roughly a 5.7 percent advantage over Opus 4.8, which corresponds to an index score near 61.4 when adjusted to the same scale.

Anthropic's update improves several internal failure modes cited in prior model generations, notably around multi-step reasoning and some context management scenarios. The gains are statistically measurable on the AAII suite but are not uniform across categories: in language understanding and throughput-sensitive tasks, the difference versus Opus is smaller or negligible. For practitioners choosing a model, that nuance matters more than the headline composite score.

Price and value

The financial trade-offs are stark. Pricing published alongside the Fable 5 rollout places its effective inference cost at roughly two times what Opus 4.8 charges for equivalent latency and context settings. Anthropic billed the upgrade as higher-end, aimed at customers prioritizing top-tier accuracy on the AAII metrics, while Opus 4.8 remains positioned as the more cost-efficient option for broader utility.

That pricing delta amplifies questions about marginal utility. A 5.7 percent average performance improvement layered over a near-doubling of price produces widely different value equations depending on workload. Teams running scale-critical, high-stakes reasoning workloads may view the premium as justifiable. For volume-driven tasks, the cost per performance improvement often favors the lower-priced Opus model.

Why it matters

The release sharpens the split between headline leaderboard performance and cost-effectiveness in production AI choices. Buyers and platform operators will need to weigh modest index gains against substantially higher operating costs, driving more focused procurement decisions by use case. The result underlines that the latest model lead on benchmarks does not automatically translate into a universal upgrade path for all customers.