Claude Fable 5 Is Here: Benchmarks, Pricing & When to Use It
We pitted Fable 5, Opus 4.8, and Sonnet 4.6 against an intentionally vulnerable app — then broke down what the new flagship model actually costs.
Putting Three Models to the Security Test
To see how fast (and how thoroughly) each model can spot real security issues, we pointed Sonnet 4.6, Opus 4.8, and the newly released Fable 5 at the same intentionally insecure app and gave each one the identical prompt:
"Review the code and list every security vulnerability you can find with severity and CVE/OWASP reference."
The app had three deliberate holes baked in: SQL injection (raw query parameters concatenated directly into SQL strings), broken authentication (no rate limiting or brute-force protection, a hardcoded JWT signing secret, and tokens issued with no expiry), and command injection (raw query parameters passed straight into an exec call).
Security Scan Results
| Model | Time Taken | Critical | High | Medium | Low | Total |
|---|---|---|---|---|---|---|
| Fable 5 | ~1m 1s | 2 | 2 | 3 | 2 | 9 |
| Opus 4.8 | 39s | 2 | 3 | 2 | 1 | 8 |
| Sonnet 4.6 (high effort) | 1m 35s | 2 | 4 | 4 | 3 | 13 |
All three models caught both planted critical issues — the SQL injection and the command injection. On the "High" severity findings, every model flagged the hardcoded JWT secret and the plaintext password storage, and both Opus and Sonnet went a step further by also catching the missing-authentication issue (CWE-862, OWASP A01).
The standout was Sonnet 4.6: it independently reached for npm audit, something neither Opus nor Fable did on their own. That extra step surfaced a known SQLite dependency CVE the other two models missed entirely — explaining why Sonnet found the most issues, even though it took the longest.
What Is Claude Fable 5?
Released on June 9, 2026 — just two days before this test — Fable 5 is Anthropic's newest "Mythos-class" frontier model, the production successor to the Mythos preview shown to the public a few months earlier. It's built for long-running agentic work: it can run for multiple days, spin up sub-agents, and check whether those sub-agents have finished their work.
Fable 5 is also notably better at parsing complex documents — diagrams, charts, and nested tables inside PDFs and spreadsheets — which makes it a strong fit for document-heavy work in finance, legal, analytics, and architecture.
Built-in Guardrails
Fable 5 ships with stronger guardrails around cybersecurity and biology topics. If a request trips one of these guardrails, it's automatically rerouted to Opus 4.8 — and billed at Opus 4.8 rates, not Fable 5's much higher pricing.
Benchmark Highlights
| Benchmark | Fable 5 | Mythos Preview | Opus 4.8 |
|---|---|---|---|
| Agentic coding | 80.3% | 77.8% | ~58% |
| Multi-disciplinary reasoning (no tools) | 59% | — | — |
| Multi-disciplinary reasoning (with tools) | 64.5% | — | — |
| Cybersecurity | 78% | — | ~40% |
On agentic coding, Fable 5's 80.3% beats the Mythos preview's 77.8%, and is roughly 22 points ahead of where Opus-class and GPT-5.5-class models land (~58%). On multi-disciplinary reasoning, Fable 5 also leads Gemini (44% / 59%) and GPT-5.5 (41% / 52%) both with and without tool use. The one place the older Mythos preview still has the edge: computer use.
Pricing: Fable 5 vs Opus 4.8 vs Sonnet 4.6
Per million tokens:
| Model | Input | Output |
|---|---|---|
| Fable 5 | $10 | $50 |
| Opus 4.8 | $5 | $25 |
| Sonnet 4.6 | $3 | $15 |
- 1-hour prompt caching on Fable 5 doubles the input rate.
- Fable 5 burns through usage roughly 2x faster than Opus, so Pro/Max subscribers will hit rate limits sooner.
- It's included free for Pro and Max plans only until June 22, 2026 — after that it switches to usage-based billing via API key.
- Fable 5 is also available through OpenRouter as
anthropic/claude-fable-5, billed the same as direct Anthropic API access.
Which Model Should You Use?
- Daily development & getting started: Sonnet 4.6.
- Architecture, planning & critical implementation: Opus 4.8.
- Stuck on a tough bug, even with Opus 4.8 at max effort: switch to Fable 5 with
/model.
Recommended Deep Dives
Read Our Blogs
Best Screen Recorder Tools: What to Look For in 2026
What makes a screen recorder good in 2026: recording quality, AI transcription, editing, and privacy — plus a checklist before you choose one.
How to Get Your First 10 Paying Users: A Step-by-Step Playbook for B2B and B2C SaaS Founders
Why the first ten matter more than the next ten thousand - and exactly how to find them.