Claude Fable 5: Anthropic's June 2026 Long-Horizon Model, Now Legacy

What Claude Fable 5 is

Claude Fable 5 was released on 9 June 2026 as Anthropic's model for long-horizon agentic work, with always-on adaptive thinking and a 1 million token context window. On LMArena's text leaderboard Claude Fable 5 has 1504 Elo (rank 6, high effort); by category, 1550 Elo in Coding (rank 4, high effort), 1520 Elo in Math (rank 6, high effort), 1550 Elo in Expert (rank 2, high effort) and 1501 Elo in Creative Writing (rank 4, high effort). It takes text and image input, returns up to 128,000 output tokens, and is priced at $10 per million input tokens and $50 per million output. Extended reasoning as a separate mode does not apply here; adaptive thinking replaces it and runs continuously.

Pricing, caching and throughput

Input is $10 per million tokens and output $50 — identical to Fable 5.1, so there is no cost saving in staying on the older model. There is a meaningful caching difference in the other direction: a cache hit on Fable 5 costs $1 per million tokens, the standard 10% of base input, where Fable 5.1 charges 2.5%. The Batch API takes 50% off input and output. Anthropic charge no long-context surcharge: the full 1 million token window bills at standard pricing. On throughput, it generates 70.1 tokens per second with a 69.4 second time to first token, and Anthropic's measurements put it at roughly 35,565 output tokens and 319 seconds per task. It supports web search, code execution, function calling and structured output, and is available through AWS Bedrock, Google Cloud and Microsoft Foundry.

Fable 5 and Fable 5.1

List prices are identical: $10 per million input tokens and $50 output. A cache hit costs $1 per million on Fable 5 (10% of base input) and 2.5% of base input on Fable 5.1. On LMArena, Fable 5.1 has 1504 Elo (rank 5, max effort). Anthropic's documentation directs most workloads to Opus 5.5, which lists at $4 and $20.

Status

Anthropic list Fable 5 as legacy following Fable 5.1 on 1 September 2026; it remains available, and the models overview groups it with the other legacy releases, Opus 5, Opus 4.8 and Opus 4.7. Opus 5.5 lists at $4 per million input and $20 output, with cache reads at $0.20 per million. There is no ongoing free tier, only a small trial credit for new accounts.

Cache writes and billing

A 5-minute cache write on Fable 5 costs $12.50 per million tokens and a 1-hour cache write $20; a cache read costs $1. Anthropic's pricing documentation states that caching pays off after one cache read for the 5-minute duration and after two reads for the 1-hour duration. US-only inference via the inference_geo parameter applies a 1.1x multiplier to all token categories. On Claude Platform on AWS and Claude in Microsoft Foundry, usage is billed through the marketplace in Claude Consumption Units at $0.01 each, converted from the standard per-token rates. Fable 5 uses the newer tokenizer introduced with Claude 4.7, which Anthropic say produces approximately 30% more tokens for the same text.

Official Resources