Claude Fable 5 Is Now Live on Tokensmart: Anthropic’s First Public Mythos-Class Model, at 30% of the Official Price

TL;DR

Zero changes to existing code: switch the model field to claude-fable-5 and you are running. Live prices on the pricing page.

What Fable 5 is

On June 9, Anthropic released Claude Fable 5 alongside Claude Mythos 5. The Mythos line had previously been limited to 150+ trusted organizations (hardening cyber defenses in finance, software, and healthcare). Fable 5 is the first Mythos-class model everyone can use — the same base model as Mythos 5, with safeguards tuned for general use.

According to Anthropic, Fable 5 is state-of-the-art on nearly every tested capability benchmark, and the longer and more complex the task, the larger its lead — long-horizon coding, asynchronous multi-step workflows, and very long document analysis are exactly its home turf.

Benchmark numbers worth knowing

• Software engineering: top of Cognition's FrontierCode eval; Stripe reported compressing months of engineering work into days
• Knowledge work: top score on Hebbia's Finance Benchmark; aced IMC's trading-analysis evaluations
• Vision: new state-of-the-art — completed Pokémon FireRed using only raw screenshots; markedly better at diagrams, tables, and PDFs
• Persistent memory: memory improves its performance 3× more than it does for Opus 4.8
• Analytics benchmarks: first model to break 90% (Opus 4.8 sits about 10 points lower)

Pricing: a flat 30% of official

Following our transparent pricing principle, platform price = official price × discount rate, and Fable 5's rate is 0.3:

• Billed on actual token usage — no subscription, no monthly fee, no minimum top-up
• Prompt caching works automatically; cache_read_tokens is listed separately in your API logs, never double-charged
• Every call's exact cost is itemized in the logs

Why the API route makes sense right now

Anthropic's own subscription plans (Pro / Max / Team / Enterprise) include Fable 5 for free only through June 22; after that it moves to usage-credit rationing — capacity is tight, by Anthropic's own admission. The API route is pay-as-you-go and instant, and Tokensmart takes the unit price down to 30% on top of that:

• No fighting for subscription quota, no waitlist
• Pay only for what you use — agent runs and batch jobs become precisely costable
• Existing keys and existing code work unchanged

How to get started

Three lines of config with the native Anthropic SDK:

import anthropic

client = anthropic.Anthropic(
    api_key="pk_live_...",
    base_url="https://api.tokensmart.ai",
)

msg = client.messages.create(
    model="claude-fable-5",
    max_tokens=4096,
    messages=[{"role": "user", "content": "Hello, Fable 5!"}],
)
print(msg.content[0].text)

Tools like Claude Code work the same way — point the base URL at the platform and paste your key. See the setup guide.

Verify the model is enabled for your account:

curl https://api.tokensmart.ai/v1/models \
  -H "Authorization: Bearer pk_live_..."

If claude-fable-5 shows up in the returned list, you are good to go.

Two transparency notes

• High-risk requests auto-fall back: per Anthropic's official mechanism, requests touching high-risk areas (offensive cyber, biology/chemistry, model distillation) are caught by classifiers and automatically fall back to Opus 4.8 (not refused). Anthropic's published figure: fewer than 5% of sessions on average. This is model-side behavior — identical through every channel
• Data retention: Anthropic applies a 30-day retention policy to all Mythos-class traffic (not used for training). That is upstream policy; we pass it on as-is

Also: claude-opus-4-8 is on sale as well (it is Fable 5's official fallback target) — live price on the pricing page.

Try it now

1. Sign up / log in
2. Create a key on the API platform
3. Switch model to claude-fable-5 and run

Questions? Reach us anytime at support@tokensmart.ai. Have fun 🚀