Large language model
Verdict
Anthropic's 1 Sep 2026 refresh of Fable. Tied #1 on Artificial Analysis Intelligence Index v4.3 (53) with GPT-6 Astra, and #1 on Arena Agent. Same $10/$50 as Fable 5, but cache reads are $0.25 (75% cheaper). The coding and knowledge-work ceiling if you can pay frontier rates.
“The model to beat in September 2026 — if budget allows.”
Use for hard coding/research; route volume to Sonnet 5, Muse Spark 1.3, or GLM-5.3 Flash.
Training data policy: API data not used for training by default (opt-out / ZDR available)
DPDP Act (India): No India-resident inference by default. Use DPA + region controls.
Data residency: US, EU
API available globally