Longer autonomous runs
Anthropic emphasizes the longest autonomous-work posture among Claude releases so far; the more complex and longer the chain, the clearer the advantage over Opus 4.8 usually becomes.
Anthropic · Claude Fable 5 · June 2026
Claude Fable 5 — built for ultra-long, complex autonomous work
- Million-token context
- Extra-long single replies
- Longest autonomous-run posture
- Safety-classifier fallback
What is Claude Fable 5?
Claude Fable 5 is Anthropic’s Mythos-class public release from June 2026. It’s built for hours-long—or longer—autonomous knowledge work and coding: Anthropic says capability exceeds any previously public Claude. Some requests related to cybersecurity, biochemistry, or distillation may be handed to Claude Opus 4.8 by a safety classifier, with the user informed; most sessions still run fully on Fable 5. Everyday engineering primary more often stays on Claude Opus 5. On iMini Agent, pick Claude Fable 5 to use it.
- Vendor
- Anthropic
- Released
- June 2026
- Context
- 1M tokens
- Max output
- 128K tokens
- Deep thinking
- Deeper on hard jobs
- Best as
- Ultra-long, ultra-hard autonomous work
What’s new versus Claude Opus 4.8
Autonomy length, cross-domain ceiling, and long-context focus—what to check before stepping up to Fable.
Cross-domain ceiling capability
Official benchmarks lift across software engineering, knowledge work, vision, and research—suited to high single-shot investment and async wait times.
More focused under long context
Million-token materials and long threads hold goals more easily; persistent file-memory style capabilities help long jobs.
Fewer steps at matched quality
Customer feedback and coding evals show reaching similar quality often needs fewer turns and steps—suited to heavy async workloads.
Official evaluations and safety audits
Results Anthropic published in the Claude Fable 5 / Mythos 5 launch materials. Most benchmark gaps between the two models are within 1–3 points; tables take the higher score. Starred cybersecurity and biology benches are safety-intercepted, so Fable 5 tracks closer to Claude Opus 4.8 there.
Three common workflows
Claude Fable 5 fits high failure cost, async wait, and very long chains.
Ultra-long coding & repo-scale changes
For migrations, refactors, and multi-day fixes across many files. Write success criteria clearly and accept longer runtimes.
Deep knowledge work & research synthesis
For long-material reasoning, chart reading, and multi-source conclusions. Focus on checkable evidence and deliverable conclusions.
Vision-driven rebuilds & analysis
For understanding UIs from screenshots, rebuilding page structure, or reading scientific figures. Keep images and long instructions in one thread.
How to choose vs Claude Opus 5 and Claude Sonnet 5
All three are current Claude models. The differences are ceiling capability, everyday primary cost, and safety-fallback experience.
| Dimension | Claude Fable 5 | Claude Opus 5 | Claude Sonnet 5 |
|---|---|---|---|
| Context | 1M tokens | 1M tokens | 1M tokens |
| Reasoning | Deeper on hard jobs, ceiling posture | Deep thinking built in, tunable | Tunable depth tiers |
| Coding / agents | Ultra-long, ultra-hard autonomous coding | Everyday flagship primary for long-horizon coding | Scaled agents & everyday coding |
| Long tool runs | Longest autonomous-run posture | Stronger goal holding and completion | Better as a high-frequency primary |
| Speed posture | Ceiling completion quality first | Quality and stability in balance | Throughput first |
| Prefer when | Single ultra-long, ultra-hard, async jobs | New long-horizon coding and knowledge work | New high-frequency agent primary |
Using Claude Fable 5 on iMini
Free quota, safety fallback, key specs, and how it differs from Opus and Sonnet.
Use Claude Fable 5 free on iMini Agent
Million-token context for ultra-long autonomous coding and knowledge work.
Open Claude Fable 5No Anthropic API key required.