claude-fable-5

Claude Fable 5 is Anthropic's third-highest-capability model in the Claude 5 family. It is a reasoning model with a 1M+ token context window, priced at the same level as Claude Mythos 5 but with different capability characteristics — Fable 5 achieves higher measured throughput and stronger head-to-head performance in direct model comparisons. It supports vision inputs, tool use, and streaming.

Claude Fable 5 is priced at $10.00 per million input tokens and $50.00 per million output tokens — the same as Claude Mythos 5. Prompt caching reduces repeated input to $1.00/1M. At 1,000 input + 500 output tokens per call, the cost is approximately $0.035 per 1,000 API calls. Measured throughput is 68 tokens per second, with the highest latency in the Claude 5 family, reflecting deep extended reasoning. For workloads where the reasoning depth of Fable 5 is required, the latency trade-off is acceptable; for interactive or latency-sensitive applications, Claude Sonnet 5 or Claude Opus 5 are more appropriate.

Claude Fable 5 supports a 1M+ token context window, enabling analysis of very large documents and codebases. Its 139-second median TTFT is the highest in the Claude 5 family, indicating the deepest reasoning process — tasks that require the most extended chain-of-thought benefit most from Fable 5's architecture. Maximum output is 65,536 tokens. The high Elo rating of 1506.49 (highest among Claude 5 models with a measured Elo) suggests it performs particularly well in head-to-head comparisons against other frontier models, even when its overall BenchAlign score is marginally below Mythos 5 and Opus 5.

Claude Fable 5 is best suited to tasks requiring the deepest reasoning depth: frontier mathematics, complex multi-step scientific analysis, advanced competitive programming, and long-horizon planning tasks where the quality of the reasoning process matters as much as the final answer. Its high Elo rating makes it a strong choice for tasks framed as direct comparisons or evaluations. For production inference at scale, the 139-second median latency makes it unsuitable for interactive workloads — it is best deployed in async batch pipelines where quality is the sole optimization target and latency is not a constraint.

Claude Fable 5 occupies an interesting position in the Claude 5 family: same price as Mythos 5, but with higher measured throughput and stronger head-to-head performance in direct comparisons. The choice between Mythos 5 and Fable 5 depends on the task: for head-to-head comparison tasks, Fable 5's higher throughput and direct-comparison strength may be preferable; for overall capability breadth, Mythos 5 is the better signal. Claude Opus 5 delivers near-identical benchmark performance at half the price and with lower latency — making it the better default for most frontier-tier production workloads.