Claude Fable 5: Anthropic's New Mythos-Class Model (Benchmarks, Pricing & What's New)

On June 9, 2026, Anthropic released Claude Fable 5, the first model in a new “Mythos-class” tier that sits above the Opus line. It landed just days after Anthropic publicly warned that frontier AI is becoming dangerously capable — and the model ships with the most aggressive safety scaffolding the company has put on a general release. This guide covers what Fable 5 actually is, the benchmark numbers circulating so far, pricing, availability, and how it stacks up against Opus 4.8, GPT-5.5, and Gemini 3.5.

A note on sourcing: Anthropic's announcement lists the evaluations it ran (Cognition FrontierCode, Hebbia Finance, CursorBench, FrontierBench) but has not published the raw benchmark numbers on its model page. Every score below is from independent third-party testing and is attributed as such. Treat the figures as early, unofficial signals until Anthropic's system card numbers are public.

What is Claude Fable 5?

Claude Fable 5 is Anthropic's most capable publicly available model to date. Anthropic describes the new “Mythos-class” tier as exceeding the capability of anything it has previously released for general use. There are two siblings:

• Claude Fable 5 — the public model, available via the API (claude-fable-5), the Claude apps, and Amazon Bedrock.
• Claude Mythos 5 — the same underlying model with safeguards lifted, available only in a limited program (“Project Glasswing”) for infrastructure providers and vetted cybersecurity researchers. It is not a general release.

So when people say “the new Mythos model,” the one developers can actually call is Fable 5.

When was Claude Fable 5 released?

Fable 5 was announced and made available on June 9, 2026. It is available immediately through the Claude API and Amazon Bedrock, and is included free for Claude Pro, Max, Team, and Enterprise users during an introductory window from June 9 through June 22, 2026. After June 23, usage draws on credits/usage-based billing.

What's new vs Claude Opus 4.8?

Opus 4.8 (released May 28, 2026) is now the step below Fable 5. The headline differences:

• A new capability tier. Fable 5 is the first “Mythos-class” model — positioned above the Opus family rather than as an Opus point release.
• Long-horizon autonomy. Third-party coverage emphasizes that Fable 5 can run agentic coding tasks for extended periods — “days” of work — with minimal human intervention.
• Materially higher benchmarks on agentic coding and tool-augmented reasoning (see below).
• 2× the price. Fable 5 costs exactly double Opus 4.8.
• Hard safeguards that fall back to Opus 4.8. Sensitive queries are rerouted to the safer Opus 4.8 model rather than answered by Fable 5.

Claude Fable 5 benchmarks

Anthropic has not released official numbers yet. The table below aggregates figures reported by independent testers and analysts; comparison columns are drawn from the same sources. Read them as directional, not authoritative.

The standout is agentic coding: independent SWE-Bench Pro testing puts Fable 5 around 80%, roughly 11 points above Opus 4.8 and more than 20 above GPT-5.5. On GPQA Diamond, reports are contested — some put Fable 5 near 95% while GPT-5.5 is measured at 93.6%, and at least one comparison omitted Fable's number entirely, so we're leaving an exact figure out until it's confirmed.

How much does Claude Fable 5 cost?

Fable 5 is priced at exactly twice Opus 4.8:

Prompt-caching and batch rates have been reported by third parties (for example a ~90% input discount on cache hits) but are not yet confirmed on Anthropic's pricing page — verify against the official docs before budgeting around them. During the June 9–22 introductory window, Pro/Max/Team/Enterprise subscribers can use Fable 5 without extra charge.

Context window and availability

• Context window: reported at 1,000,000 tokens by third-party sources; Anthropic's model page does not state it explicitly.
• Max output: up to 128,000 tokens per response (per multiple secondary sources).
• API model ID: claude-fable-5.
• Where to run it: Claude API and the Claude apps; Amazon Bedrock (confirmed by AWS). Availability on Google Vertex AI and Microsoft Foundry has been reported by third parties but isn't AWS/Anthropic-confirmed.

Claude Fable 5 vs GPT-5.5 vs Gemini 3.5

On the numbers circulating today, Fable 5 leads agentic coding and tool-augmented reasoning by a wide margin — SWE-Bench Pro 80.3% versus GPT-5.5's 58.6% and Gemini 3.1 Pro's 54.2%. The trade-offs:

• GPT-5.5 remains competitive on knowledge benchmarks (it's measured slightly ahead on GPQA Diamond) and is cheaper.
• Gemini 3.5 Flash is dramatically faster and cheaper for high-volume agent loops, trading peak capability for throughput.
• Fable 5 is the capability leader for hard, long-running coding and agent work — but at $10/$50 it is the most expensive of the three, so it's a “use it where it pays for itself” model rather than a default.

The safety story: why some answers come from Opus 4.8

Fable 5 shipped with unusually hard guardrails. Anthropic says queries touching cybersecurity exploitation, bio/chem weapons, and model-distillation are blocked at the model level and rerouted to the safer Opus 4.8 — a behavior it expects to trigger in under 5% of sessions. The company also reports no universal jailbreaks found across more than 1,000 hours of red-teaming/bug-bounty, and a mandatory 30-day traffic-retention policy for the model. Anthropic points to a separate system card and risk report for the full details; the specific ASL safety level is not stated in the launch announcement, so we're not quoting one here.

How to use Claude Fable 5

If you already call the Claude API, switching is a one-line model change:

from anthropic import Anthropic

client = Anthropic()
msg = client.messages.create(
    model="claude-fable-5",
    max_tokens=8192,
    messages=[
        {"role": "user", "content": "Plan and implement the refactor in this repo."}
    ],
)
print(msg.content[0].text)

It's also selectable in the Claude apps (free through June 22 for paid tiers) and in Claude Code for agentic coding sessions. On Bedrock, use the corresponding Bedrock model identifier rather than the raw API ID.

Should you use Fable 5?

For most day-to-day work, Opus 4.8 or a faster model like Gemini 3.5 Flash is the better cost/throughput choice. Reach for Fable 5 when the task is genuinely hard and long-horizon — multi-hour autonomous refactors, complex agent pipelines, or problems where Opus 4.8 plateaus — and where the output quality justifies paying double. Because the benchmark numbers are still third-party and the model is one day old, validate it on your own tasks before committing budget.

FAQ

Is Claude Fable 5 better than Claude Opus 4.8?

On the third-party benchmarks available so far, yes — notably on agentic coding, where independent SWE-Bench Pro testing puts Fable 5 around 80% versus roughly 69% for Opus 4.8. It also costs twice as much, and its hardest safety-blocked queries are rerouted to Opus 4.8.

How much does Claude Fable 5 cost?

$10 per million input tokens and $50 per million output tokens — exactly double Opus 4.8. Paid Claude subscribers get free access during the June 9–22, 2026 introductory window.

What's the difference between Fable 5 and Mythos 5?

They're the same underlying “Mythos-class” model. Fable 5 is the public version with safeguards on; Mythos 5 is a restricted, safeguards-lifted version available only to vetted infrastructure providers and cybersecurity researchers through a limited program.

What is the Claude Fable 5 API model ID?

It's claude-fable-5. The model is available via the Claude API, the Claude apps, and Amazon Bedrock.

What is the Claude Fable 5 context window?

Third-party sources report a 1,000,000-token context window with up to 128,000 output tokens per response. Anthropic's model page does not state the context size explicitly, so confirm against the official docs.

Why are some of my Fable 5 answers handled by Opus 4.8?

Anthropic blocks sensitive categories (cyber-exploitation, bio/chem, distillation) at the model level and reroutes them to the safer Opus 4.8. It expects this to affect under 5% of sessions.

Is Claude Fable 5 free?

It's included for Claude Pro, Max, Team, and Enterprise subscribers from June 9 to June 22, 2026. After that, usage is billed via credits/usage-based pricing.