Claude Fable 5 went live on June 9, and unlike some frontier releases it is not waitlisted. You can call it right now. This guide covers what it costs, who gets it included, and a practical answer to the question that actually matters: when is it the right pick?
What it costs
The full Claude lineup as of today, per million tokens, input and output:
- Haiku 4.5: $1 / $5, the high-volume workhorse.
- Sonnet 4.6: $3 / $15, the production default.
- Opus 4.8: $5 / $25, the previous flagship, released May 28.
- Fable 5: $10 / $50, the first Mythos-class model.
Fable 5 lands at exactly double Opus 4.8, per Anthropic's announcement. For context, security partners have paid $25/$125 for the restricted Mythos Preview since April, so the public tier arrives at less than half the partner rate. Launch coverage reports prompt-caching discounts carry over as on other Claude models; check the current platform docs before you budget around that.
The June 22 window
Here is the detail many people will miss: Fable 5 is included on Pro, Max, Team, and seat-based Enterprise plans through June 22. After that, subscription users need usage credits to keep calling it. If you have a Claude subscription, you have roughly two weeks to test the model against your real workload at no extra cost. Treat it as an evaluation window with a deadline.
When to pick Fable 5 over Opus or Sonnet
Double the price needs a reason. A simple decision guide:
- Pick Fable 5 for long autonomous runs, multi-hour agentic workflows, codebase-wide refactors and migrations, and analysis that chains many steps. This is where it pulls away: 80.3% on SWE-Bench Pro versus 69.2% for Opus 4.8, and runs that finish 25 to 30% faster with fewer turns. The detailed numbers are in our three-way benchmark comparison.
- Stay on Opus 4.8 for short interactive coding, drafting, and review. It keeps most of the capability at half the price. It is also the model that answers anyway when Fable 5's safety classifiers decline a query, so for security-research-adjacent topics, starting on Opus skips a reroute.
- Stay on Sonnet 4.6 or Haiku 4.5 for high-volume production pipelines where per-token cost dominates and tasks are well-scoped.
Before you switch: two operational notes
First, Mythos-class traffic carries a 30-day data retention policy, which matters if your compliance posture assumes zero-retention API calls. Second, a small share of sessions, under 5% on Anthropic's numbers, get rerouted to Opus 4.8 by the safety layer, and the classifiers are deliberately cautious. If your domain sits near cybersecurity or the life sciences, expect an occasional benign request to trip them. The mechanics are explained in our guardrails deep dive.
Getting started
On the API, the model ID is claude-fable-5. On claude.ai, it appears in the model picker on included plans. One unconfirmed item as of launch day: Anthropic has not published the context window size, so do not architect around an assumed number until the docs say so. For the bigger picture of what this model is and where it came from, start with the full explainer.