Claude Fable 5.1

Claude Fable 5.1 (1 September 2026) is Anthropic's flagship for coding and knowledge work — same $10/$50 sticker as Fable 5, but cache reads cut 75% to $0.25 per million, which is what actually lowers the bill on agentic work.

Best for

• Long-running agentic coding and multi-step research
• Knowledge work: documents, spreadsheets and slides
• Codebase-scale context — 1M tokens in, 128K out
• Workloads where the same context is re-read hundreds of times

How it compares — and the India angle

• Anthropic's own documentation says to start with Opus 5 ($5/$25) and reach for Fable 5.1 only when Opus at higher effort still falls short — it is the top of the range, not the default.
• The headline saving is cache reads at $0.25/1M, a quarter of GPT-6 Astra's $1.00 and only 25% above Sonnet 5's, despite a base input price five times Sonnet's. For Indian teams running persistent agents, that is the line that matters.
• Honest caveat: Artificial Analysis found that at maximum effort Fable 5.1 uses ~1.7x the output tokens of Fable 5, pushing cost per task up ~20%. The 25-45% saving is for typical and cache-heavy work, not the top effort setting.
• Its lowest effort setting beat Fable 5's highest on agentic science (~26% vs 24.7%) at roughly a quarter of the cost per task — try dialling effort down before paying for a bigger model.