Disclosure: Some links below are affiliate links. If you buy through them, this site earns a small commission at no extra cost. Editorial recommendations are never influenced by affiliate rates.
In this article
Anthropic released Claude Fable 5 on 9 June 2026, a Mythos-class AI model that was previously too dangerous to ship to the public. Here's what's actually different, what the guardrails mean in practice, and whether the 2x price jump over Opus 4.8 is justified.
What is Claude Fable 5?
Fable 5 is the publicly accessible version of Anthropic's Mythos model. Mythos launched in April 2026 and was kept locked behind a restricted programme called Project Glasswing because, in Anthropic's own words, it could identify and exploit vulnerabilities "in every major operating system and every major web browser." That's not a product you can hand to everyone.
Fable 5 solves this by attaching new classifiers that block responses in high-risk areas. Ask about exploiting a CVE or synthesising a dangerous compound, and Fable 5 doesn't just refuse. It hands the query off to Claude Opus 4.8, which answers with the usual safety filters. You get a response either way; the Mythos-class reasoning just doesn't apply to the dangerous parts.
What makes it better than Opus 4.8?
The headline number is 10%+ higher on select benchmarks versus Opus 4.8. On SWE-bench Verified, the Mythos preview posted 93.9%, compared to around 80.8% for Opus 4.6. Coding and knowledge work tasks are where the gap is most visible.
Anthropic describes the performance jump as a "significant step" rather than an incremental update. Dianne Penn, Anthropic's Head of Product Management for Research, put it plainly: higher intelligence means higher ROI per task, even at higher cost.
Real-world performance differences that matter:
- Multi-step software engineering tasks: Mythos-class models handle longer chains of reasoning before losing the thread
- Complex analysis and research synthesis: Fewer hallucinations on detailed factual queries
- Agentic tasks: Better at planning and self-correction in autonomous workflows
What does it cost?
| Claude Opus 4.8 | Claude Fable 5 | |
|---|---|---|
| Input (per million tokens) | ~$5 | $10 |
| Output (per million tokens) | ~$25 | $50 |
| Context window | 200K | 200K |
| Availability | API, Claude.ai Pro | API, Claude.ai Enterprise + Pro |
That's a straight 2x price increase. For most individual use cases, you won't notice the jump. For companies running Fable 5 at scale through the API, this adds up quickly.
The ROI case depends on what you're using it for. If a task currently fails 1 in 5 times with Opus 4.8 and passes 9 in 10 with Fable 5, the maths work. If you're drafting marketing copy or summarising documents, pay for Opus 4.8.
Who should actually switch?
Switch to Fable 5 if:
- You're running complex agentic workflows where reasoning errors are expensive
- You need the best available coding performance for production engineering tasks
- You're already spending heavily on API calls and accuracy matters more than per-token cost
Stick with Opus 4.8 if:
- Your workloads are writing, summarisation, or customer support
- You're cost-sensitive or running at high volume
- You don't need frontier-level reasoning for everyday tasks
Fable 5 is available through Claude.ai Pro and Enterprise plans. It's not in the free tier.
The bigger picture
Anthropic filed its IPO prospectus confidentially just days before this launch. The company is currently running at a $47 billion revenue run rate and was valued at $965 billion as of June 2026, which puts it ahead of OpenAI's $852 billion valuation from March. OpenAI also filed its own IPO prospectus on 8 June 2026.
That context matters. Fable 5 is a product launch, but it's also a signal to future public investors that Anthropic can commercialise Mythos-class capability safely. The timing is deliberate.
Tip
If you want to test Fable 5 before committing to an Enterprise plan, Anthropic is rolling it out to Claude.ai Pro subscribers. Run your most demanding tasks through it for a month and compare accuracy against what you're getting today.
Bottom line
Claude Fable 5 is the most capable model Anthropic has made available to the public. If your workflows involve complex coding, long-form analysis, or agentic tasks where reasoning quality directly affects output quality, the 2x price is probably justified. If you're doing simpler work, Opus 4.8 is still excellent and costs half as much.
The guardrail architecture is genuinely clever. Rather than refusing high-risk queries entirely, Fable 5 downgrades to a safer model for those cases. In practice, this means you get Mythos-level intelligence where it's safe to have it, without creating a liability.
For developers building on the Claude API, start with Opus 4.8 and upgrade to Fable 5 only once you've identified the specific tasks where the accuracy gap costs you real money.