Claude Fable 5 and Mythos 5 Release — Next-Gen Models for Long-Running Complex Tasks

Anthropic announced Claude Fable 5 and Claude Mythos 5 on June 9, 2026. Fable 5 is now generally available via API and subscription plans, delivering state-of-the-art performance on long-running, complex tasks. Mythos 5 is the safeguards-removed version for Glasswing partners. In my experience using Claude Code daily alongside Zed, this new generation significantly reduces the time required for large-scale refactoring projects such as 50M-line codebase migrations.

📑Table of Contents

1. What is Claude Fable 5 / Mythos 5 — Positioning of the Next-Gen Models
2. Pricing Changes and Plans — Comparison Table with Opus 4.8
3. Key Benchmarks and Performance on Long-Running Complex Tasks
4. Safeguard Mechanism and Differences from Mythos 5
5. Developer and API Usage Steps and Limitations
6. Frequently Asked Questions (FAQ)
7. Conclusion

---

What is Claude Fable 5 / Mythos 5 — Positioning of the Next-Gen Models

Claude Fable 5 is positioned as the successor to Opus 4.8 in the Mythos class, with context length expanded to 1M tokens. It achieves SOTA results on long-horizon agent tasks and massive refactoring workloads. Mythos 5 shares the same base model but removes safeguards for trusted partners in the cyber domain via Project Glasswing. From my practical use of Claude Code within the Zed editor, this expanded context handling proves extremely valuable for real-world development workflows.

---

Pricing Changes and Plans — Comparison Table with Opus 4.8

Pricing has been reduced to $10 per million input tokens and $50 per million output tokens, roughly half of the previous generation. Prompt caching offers a 90% discount on input tokens. On claude.ai paid plans, usage counts as 2x. As someone who combines ChatGPT Pro ($100) and Claude Max ($200) plans, the lowered API costs make Fable 5 much more accessible for production workloads.

Item	Fable 5	Opus 4.8
Input (per M tokens)	$10	$15
Output (per M tokens)	$50	$75
Context Window	1M	200K
Prompt Caching	90% off	N/A

Source: Anthropic official announcement and TrueFoundry (June 2026)

---

Key Benchmarks and Performance on Long-Running Complex Tasks

Fable 5 scored 80.3% on SWE-Bench Pro, significantly ahead of Opus 4.8 at 69.2%. On FrontierCode Diamond it reached 29.3% compared to Opus 4.8’s 13.4%. A real-world example includes completing a 50M-line Ruby codebase migration for Stripe in a single day. In my Claude Code usage for extended design sessions, this sustained performance represents a major advantage.

Benchmark	Fable 5	Opus 4.8	GPT-5.5
SWE-Bench Pro	80.3%	69.2%	58.6%
FrontierCode Diamond	29.3%	13.4%	–
Legal	13.3%	–	2.1%

Source: Vellum and TrueFoundry (June 2026)

---

Safeguard Mechanism and Differences from Mythos 5

Fable 5 includes safety classifiers that route flagged queries (cyber, bio/chem, distillation) to Opus 4.8, affecting less than 5% of sessions. No additional charges apply for reroutes. Mythos 5 is the fully de-restricted version available only to Glasswing partners. When calling the API directly, understanding these mechanisms helps maintain both safety and cost efficiency.

---

Developer and API Usage Steps and Limitations

The API model identifier is “claude-fable-5”. Access is available through AWS Bedrock, Google Vertex AI, and TrueFoundry Gateway. On claude.ai, usage counts double. From my experience, for batch-style or detached IDE work, direct API calls offer better cost control than repeated IDE sessions. Note that access is currently suspended due to US government export controls while Anthropic works on restoration.

---

Frequently Asked Questions (FAQ)

Q1: What are the main differences between Fable 5 and Opus 4.8?

Fable 5 expands context to 1M tokens and achieves 80.3% on SWE-Bench Pro. Pricing is significantly lower, making it ideal for long-running tasks.

Q2: Is a 50M-line codebase migration really possible in one day?

Yes, confirmed in official announcements and third-party benchmarks. Stripe completed a 50M-line Ruby migration in a single day.

Q3: Can general users access Mythos 5?

No. Mythos 5 is limited to Glasswing partners. General users should use Fable 5.

Q4: Has the price really been halved? What about caching effects?

Input is $10/M and output $50/M — roughly half the previous generation. Prompt caching adds a further 90% discount on inputs.

Q5: How often are queries blocked by safeguards?

Less than 5% of flagged queries are routed to Opus 4.8. No extra charges apply for fallbacks.

Q6: What is the exact model name for Fable 5 via API?

The model name is “claude-fable-5”. It is also available via Bedrock and Vertex.

---

Conclusion

Claude Fable 5 represents a major leap for long-horizon and large-scale tasks. Combined with reduced pricing, it expands practical possibilities for developers like myself who rely on Claude Code daily. When using the API, pay close attention to safeguard behavior and cost management. Monitor official channels for the latest access status.

As a next step, check the official Anthropic site for updates, obtain an API key, and start testing with smaller tasks to evaluate fit for your workflow.

Related articles: Claude Opus 4.8 リリース：Claude CodeのDynamic Workflowsと高速・低コスト化を解説、DynatraceがAI Coding Agent監視を拡張：Claude Code・Gemini CLI・Codex CLIをOpenTelemetryで可視化。