Claude Fable 5 & Mythos 5: Anthropic's Most Powerful Models Go Public
On June 9, 2026, Anthropic released Claude Fable 5 — its first "Mythos-class" model made safe for general use — alongside a restricted Claude Mythos 5 for vetted cyberdefenders. Here's what launched, what it costs, and what it means for developers building real apps.
The 30-Second Version
Fable 5 is Anthropic's most capable generally available model ever — state-of-the-art on nearly every benchmark it was tested on, spanning software engineering, vision, long-context reasoning, and scientific research.
Mythos 5 is the same underlying model with some safeguards lifted, available only to a small set of vetted cyberdefenders and infrastructure providers through Project Glasswing.
Pricing: $10 per million input tokens / $50 per million output tokens — less than half the price of the earlier Mythos Preview.
Safety: sensitive queries (cyber, bio/chem) automatically fall back to Claude Opus 4.8. Anthropic reports this triggers in fewer than 5% of sessions.
Fable 5 vs Mythos 5: At a Glance
Claude Fable 5
General release · API id claude-fable-5
- Available now via the Claude API
- On Pro, Max, Team & Enterprise plans (staged rollout)
- Also on Amazon Bedrock
- Built-in safeguards; falls back to Opus 4.8 on risky queries
- $10 / $50 per M tokens (input / output)
Claude Mythos 5
Restricted access
- Same underlying model, some safeguards lifted
- Limited to vetted cyberdefenders & infra providers
- Deployed through Project Glasswing
- Upgrade to the earlier Claude Mythos Preview
- Separate biology-access program for select researchers
What is a "Mythos-class" model?
Mythos is the codename Anthropic gave to a generation of models that represented what it called a "step change" in capability earlier in 2026. The catch: these models were so good at finding and exploiting software vulnerabilities that Anthropic initially decided not to release them publicly, offering access only to select entities through a vetted program. Unfettered access, the reasoning went, would hand cybercriminals a tool to discover and weaponize flaws at scale.
Fable 5 is the compromise: the same frontier model, wrapped in safeguards that make it safe enough for general availability. It is the first Mythos-class model the public can actually use.
What Fable 5 can do
Anthropic claims state-of-the-art results on nearly every benchmark it tested. The numbers that matter most to builders:
Software engineering
Tops Cognition's FrontierCode evaluation. Stripe reported using it to complete a 50-million-line Ruby migration in a single day — work it estimated would have taken a team two months. Cursor's CEO said it "opened up a class of long-horizon problems that were out of reach for earlier models."
Vision
Described as state-of-the-art for vision tasks — extracting precise numbers from scientific figures and rebuilding working web apps directly from screenshots.
Long-context reasoning
Maintains focus across millions of tokens. Anthropic reports a 3× improvement in complex games when the model is given persistent memory across a session.
Knowledge & research work
Highest scores on Hebbia's Finance Benchmark and IMC trading-analysis evals. First model to clear 90% on certain analytics benchmarks — a 10-point jump over Opus 4.8.
Pricing
Both Fable 5 and Mythos 5 are priced identically, and notably cheaper than the earlier Mythos Preview:
| Model | Input (per M tokens) | Output (per M tokens) | Notes |
|---|---|---|---|
| Claude Fable 5 | $10 | $50 | General availability |
| Claude Mythos 5 | $10 | $50 | Restricted (Project Glasswing) |
For context, that's frontier-tier pricing — meaningfully more than a workhorse model like Claude Haiku 4.5. For most app features (chat, summarization, classification, structured extraction) a cheaper model is still the right default. Reach for Fable 5 when a task genuinely needs its reasoning ceiling: large refactors, multi-file architectural changes, deep research, or anything that hits a wall on smaller models.
The safety model: an automatic Opus 4.8 fallback
How the safeguards work
Fable 5 ships with a classifier-based system. When a request touches a high-risk area, the response is generated by Claude Opus 4.8 instead of Fable 5. Anthropic says more than 95% of sessions involve no fallback at all. There are three guarded areas:
- Cybersecurity — blocks exploitation queries and agentic hacking. External red-teaming reported zero successful harmful single-turn requests across 30 public jailbreak techniques.
- Biology & chemistry — broadly blocks dual-use research queries.
- Distillation — prevents extracting the model's capabilities to train unauthorized models.
Mythos-class traffic also carries a 30-day data-retention requirement, with logged human access and automatic deletion; the data is not used for training.
The practical implication for developers: if your product touches security tooling or biotech, expect some responses to come from Opus 4.8 rather than Fable 5. For typical SaaS, content, and app-building workloads, you'll rarely hit the fallback.
Availability & rollout
- Claude API — available immediately as
claude-fable-5. - Subscription plans — included on Pro, Max, Team, and Enterprise through June 22, 2026. After June 23, usage credits may be required if capacity is constrained, with full restoration planned as capacity allows.
- Amazon Bedrock — Fable 5 is available on AWS at launch.
- Mythos 5 — restricted to Project Glasswing partners, with a separate biology-access program (cyber safeguards remain) launching for select researchers.
What this means for your stack
You don't need to put Fable 5 behind every feature. A sensible pattern in 2026 is tiered model routing:
Cheap & fast tier
Claude Haiku 4.5 for high-volume, well-scoped tasks — chat, tagging, extraction, structured JSON.
Balanced tier
Claude Sonnet 4.6 / Opus 4.8 for general reasoning, agents, and most user-facing intelligence.
Frontier tier
Claude Fable 5 for the hardest long-horizon problems — large refactors, deep research, complex multi-step agents.
Route the bulk of your traffic to a cheaper model and escalate to Fable 5 only when a task warrants it. That keeps costs sane while still giving you frontier capability where it counts.
Frequently Asked Questions
Is Fable 5 better than Opus 4.8?
On capability benchmarks, yes — Anthropic positions Fable 5 as its most capable generally available model, beating Opus 4.8 on the tests it published (including a 10-point jump on certain analytics benchmarks). Opus 4.8 remains the model that handles safeguarded, high-risk queries.
What's the difference between Fable 5 and Mythos 5?
They're the same underlying model. Fable 5 is the public version with safeguards on; Mythos 5 has some safeguards lifted and is restricted to vetted cyberdefenders and infrastructure providers via Project Glasswing.
How much does Fable 5 cost?
$10 per million input tokens and $50 per million output tokens — under half the price of the earlier Mythos Preview, but still frontier-tier compared to workhorse models like Haiku.
Will my app ever get an Opus 4.8 response instead of Fable 5?
Only if a request trips one of the safeguard classifiers (cyber, bio/chem, or distillation). Anthropic says that happens in fewer than 5% of sessions, so most builders won't notice it.
Can I use Fable 5 on my Claude subscription?
Yes — it's rolling out to Pro, Max, Team, and Enterprise plans, plus the Claude API and Amazon Bedrock. The plan rollout is staged, and usage credits may apply after June 23, 2026 if capacity is constrained.
Important Disclaimer
Information Accuracy: This article is based on Anthropic's launch announcement and public reporting as of June 10, 2026. Model capabilities, pricing, availability, and safeguards may change. Verify current details on Anthropic's official site before making decisions.
Trademarks: Claude, Fable, Mythos, Opus, and Anthropic are trademarks of Anthropic. This site is not affiliated with, endorsed by, or sponsored by Anthropic. Names and logos are used for informational purposes only.
No Guarantees: Recommendations here are general information, not professional advice. Evaluate models against your own requirements.
Build Your AI App Stack
Picking a model is one piece of the puzzle. Generate a complete tech stack — AI models, hosting, database, auth, and more — tailored to your budget and app type.
Related Articles
The top AI coding assistants compared
Cursor vs Claude Code 2026Detailed head-to-head comparison
Build an AI App with Next.jsNext.js, Supabase & an LLM, end to end
What is Vibe Coding?The complete guide to AI-first development
Best Tech Stack for SaaS 2026Complete guide to building modern SaaS
Solo Founder Tech Stack 2026Best tools for bootstrapped startups