Anthropic ships Claude Fable 5.1 and Mythos 5.1, reshaping long-context writing and editing economics

Anthropic launches Claude Fable 5.1 and Claude Mythos 5.1

1, positioning them as new flagships for coding and knowledge work. The two models share core specifications: a 1M token context window and a 128K output ceiling. 1 stays inside Project Glasswing for cybersecurity and life sciences work.

1 but with fewer restrictions, and reports that it designed protein binders with a hit rate near 50 per cent against a typical 10 to 15 per cent baseline. For text workflows specifically, the launch centres on longer-running drafting and research tasks that previously had to be broken across multiple sessions.

Pricing and plan access change the economics of long drafts

1 is $10 per million input tokens and $50 per million output tokens, a rate that places it above the rest of the Claude family on raw cost. 25 per million input for cached reads. 1 is briefly off the table for general users because they cannot have it.

1 is "in your Max plan, not free," because heavy usage will still burn through weekly quotas. For editors running continuous drafting pipelines, the practical question is whether per-token savings on long documents outweigh the capped weekly allowance.

Benchmark signals favour Fable 5.1 on long-running knowledge work

A weekly model round-up places Claude Fable 5 set to max reasoning with fallback at 66.04 per cent, $28.73 and 38 minutes per task, and notes that Claude Fable 5.1 set to max reasoning closed the gap further. The same issue frames the OpenAI and Anthropic race for the top overall spot as the central story of the moment. Independent user reports describe lower cache-read costs, updated safeguards, and stronger long-running task performance compared with the prior release. Together, these signals point to a model optimised for documents and research workflows where latency is less important than staying on task for the full duration of a multi-hour project.

Safeguards, detection and the writing-tooling backdrop

1 was evaluated with production safeguards enabled and scored zero on the referenced safety metric, a data point several outlets flag as material for compliance reviewers. The launch also lands against a backdrop of broader platform moves covered in adjacent reporting: OpenAI shipping GPT-6 Astra for enterprise drafting and documents, Anthropic explaining Claude's invisible text watermark, and Anthropic disclosing five blocked attempts to misuse Claude for biological-weapons research. For practitioners, the throughline is that capability gains are arriving alongside tighter watermarking and evaluation disclosures, so writing-tool rollouts now arrive with provenance and review requirements attached.

A practical way to read a generated draft against those requirements is to first clean it with a Strikethrough Text pass and then verify characters with a Special Characters Copy and Paste check before publication.

How a writer or editor can put this to work this month

1, because the Max allowance is the binding constraint rather than the listed API rate. 1 with the same prompt used on the prior model, measuring wall-clock time and accepted-task cost. 1 launch as a checkpoint for adopting watermarking-aware editing habits, since model output is increasingly expected to carry provenance signals.

For day-to-day editing support between those checkpoints, a Palindrome Checker and the Text Repeater handle the small repetitive jobs, while a Percentage Calculator makes cache-hit savings legible when reconciling invoices at the end of the week.