- 10
- June
Claude Fable 5 is Anthropic's newest AI model, released on June 9, 2026 — the first model in the Mythos family made "safe enough" to release to the public. Anthropic says Fable 5's capabilities exceed those of any model it has ever made generally available, leading on nearly every benchmark across software engineering, knowledge work, and vision — and the longer and more complex the task, the larger Fable 5's lead grows.
But that power comes with three trade-offs every organization should understand first: it is twice the price of Claude Opus 4.8 ($10/$50 per 1M tokens), it ships with safeguards that reroute answers to Opus 4.8 on high-risk topics (cyber, biology, chemistry), and it carries a 30-day data-retention policy for safety that differs from standard Claude models. This article summarizes every verified fact from Anthropic's announcement and what it means for Thai enterprises and ERP work.
Claude Fable 5 by the Numbers
Table of Contents
What is Claude Fable 5?
Claude Fable 5 is the "safe for general use" version of Anthropic's Mythos-class model — the high-capability model first unveiled in a limited program as Claude Mythos through Project Glasswing in April 2026, available only to partners such as AWS, Microsoft, Apple, and CrowdStrike. Anthropic has now wrapped Mythos-class capability in a set of safeguards and opened it to the public as Fable 5.
The headline is a clear jump in capability, especially on work that resembles real enterprise tasks — agentic coding, reading and summarizing large volumes of data, vision (images, charts, screenshots), and scientific research. Anthropic stresses that "the longer, more tool-heavy, multimodal, ambiguous, and closer to real work the task, the larger Fable 5's lead."
In short: Claude Fable 5 = the first Mythos-class model the public can use, the most capable model Anthropic has ever made generally available — traded off against twice the price of Opus 4.8, safeguards that fall back to Opus 4.8 on high-risk topics, and a 30-day data-retention policy for safety.
Fable 5 at a glance
| Item | Detail |
|---|---|
| Release date | June 9, 2026 |
| Model family | Mythos-class (public version with safeguards) |
| Pricing | $10 / 1M input tokens, $50 / 1M output tokens |
| vs Opus 4.8 | 2x the price (Opus 4.8 = $5/$25), but cheaper than the earlier Mythos Preview |
| Availability | Claude API, consumption-based Enterprise plans, AWS Bedrock, GitHub Copilot |
| Data retention | All traffic retained 30 days for safety, then deleted in nearly all cases |
| Trade-off (safeguards) | High-risk queries fall back to Opus 4.8 |
Fable 5 vs Mythos 5
Anthropic launched two names at once, and many people are confused — the truth is both are the same underlying model, differing only in how much of the safeguard layer is in place.
| Aspect | Claude Fable 5 | Claude Mythos 5 |
|---|---|---|
| Underlying model | Mythos-class (same model) | Mythos-class (same model) |
| Safeguards | Full safeguards — falls back to Opus 4.8 on high-risk topics | Safeguards selectively lifted for authorized work |
| Who can access | Public — anyone via the API | Restricted to cyberdefenders and a biology trusted-access program |
| Intended use | General enterprise work — coding, knowledge work, vision | Cyber defense, drug/protein research that needs full capability |
For most Thai organizations, Fable 5 is the relevant model; Mythos 5 is the unlocked version for specialized work. Anthropic notes that internal protein-design experts used Mythos 5 to accelerate parts of the drug-design process by about 10x, and that scientists preferred Mythos 5's hypotheses over Opus-class models roughly 80% of the time in blinded comparisons.
Benchmarks — Coding, Knowledge, Vision
Per Anthropic's published figures, Fable 5 leads Opus 4.8 in nearly every category, with an especially wide gap on agentic coding:
| Benchmark | Opus 4.8 | Fable 5 |
|---|---|---|
| SWE-bench Verified (coding) | 88.6% | 95.0% |
| SWE-bench Pro (hard coding) | 69.2% | 80.0% |
| FrontierCode Diamond (hardest tier) | 13.4 | 29.3 (more than 2x) |
| GDPval-AA (knowledge work, Elo) | 1890 | 1932 |
A concrete example Anthropic cited: Stripe reported that Fable 5 "compressed months of engineering into days," completing a migration of a 50-million-line Ruby codebase in one day versus roughly two months by hand.
Vision and long-context strengths
Anthropic calls Fable 5 "the new state-of-the-art model for tasks involving vision" — it can extract precise numbers from scientific figures and rebuild a web app's source code from a screenshot. On long-context work, when playing Slay the Spire with file-based memory, Fable 5 improved 3x more than Opus 4.8 — reflecting that the longer the task and the more context it must hold, the bigger its edge.
Safeguards — Why Some Answers Come From Opus 4.8
Because Fable 5 is so capable that it could be misused, Anthropic added three classifier-based safeguards that screen requests. If a request falls into a high-risk group, the system does not let Fable 5 answer — it automatically routes the response to Claude Opus 4.8 instead.
| Screened group | What it covers |
|---|---|
| Cybersecurity | Exploitation and agentic hacking tasks |
| Biology and chemistry | High-risk biological research |
| Distillation | Attempts to extract capabilities to train other models without authorization |
Anthropic states that more than 95% of sessions involve no fallback at all — meaning typical enterprise work is barely affected — and that external red-teamers spent over 1,000 hours attacking the safeguards without finding any "universal jailbreak."
What security teams should know
If your organization uses AI for security work — vulnerability analysis or defensive penetration testing — some requests may be silently rerouted to Opus 4.8, meaning the result for those topics equals Opus 4.8, not Fable 5. Understand this before planning to use Fable 5 for cybersecurity work.
Pricing and Availability
Fable 5 and Mythos 5 are priced identically at $10 per 1M input tokens and $50 per 1M output tokens — twice the price of Opus 4.8 ($5/$25) — but Anthropic says this is substantially cheaper than the earlier Mythos Preview at comparable or stronger performance.
| Channel | Status at launch |
|---|---|
| Claude API | Available immediately |
| Enterprise plans (consumption-based) | Available immediately |
| Subscription plans | Included at no extra cost June 9–22, 2026; then via usage credits until capacity allows permanent access |
| AWS Bedrock | Available |
| GitHub Copilot | Generally available |
The prices cited here are Anthropic's USD list prices (verified June 10, 2026). Prices and terms can change; always confirm with official sources before deciding. If your organization is considering Claude for a team, see how Team and Enterprise differ and how to order / request a quote.
How Your Data Is Handled (30 Days)
This is the part Thai organizations should read most carefully. Anthropic states that Mythos-class models (which include Fable 5) carry a policy of retaining all traffic for 30 days for safety purposes only, then deleting it after 30 days in nearly all cases.
That means Zero Data Retention (ZDR) terms that may apply to other Claude models do not apply to Fable 5, because the 30-day retention is a safety requirement specific to this model family. Organizations familiar with Claude's Zero Data Retention should understand that this point is different.
A note on PDPA and sensitive data
Before feeding personal or confidential organizational data into Fable 5, consider that the data is processed on Anthropic's cloud and retained for 30 days — unlike an on-premise ERP, where data never leaves the organization's servers. Teams subject to PDPA should set a clear policy on which data types may be used with Fable 5, and see further guidance on Claude security and data governance.
Impact on Thai Enterprises + ERP
Fable 5's stronger agentic and vision capabilities directly affect how AI can assist within ERP systems — especially long, complex tasks that require reading varied documents and screens, exactly where Fable 5 pulls ahead.
| Enterprise task | How Fable 5 helps | What to consider |
|---|---|---|
| Accounting / documents | Read invoices/receipts from images, extract precise numbers, assist reconciliation | Sensitive data retained 30 days — check PDPA first |
| Development / systems | Project-wide migration and code review (95% SWE-bench Verified) | Some cybersecurity work falls back to Opus 4.8 |
| Executive reporting | Summarize large volumes; build dashboards from images/multi-source documents | Token cost is 2x that of Opus 4.8 |
| Long-running AI Agent work | Long tasks that must hold lots of context (3x better long-context) | Budget control — long agentic runs consume many tokens |
Saeree ERP is developing an AI Assistant
Saeree ERP, by Grand Linux Solution Co., Ltd., is developing an AI Assistant to help users work with the ERP more easily — prioritizing organizational data control first, in the same direction Fable 5 is pushing AI toward more real-world work. It is currently in development and not yet generally available. To follow progress, contact our team.
Should You Use Fable 5?
| A good fit for | Consider first |
|---|---|
| Long, complex, multimodal work needing top quality | 2x the price of Opus 4.8 ($10/$50) |
| Code writing/review and project-wide migration | High-risk topics fall back to Opus 4.8 |
| Vision work — images, charts, screens, documents | Data retained 30 days — unsuitable for some sensitive data |
| Tasks Opus 4.8 genuinely can't handle | Opus 4.8 is enough and cheaper for most everyday work |
Practical advice: default to Opus 4.8 for everyday work and escalate to Fable 5 only for genuinely hard, long, or multimodal tasks where the quality is worth the higher price — because on rerouted topics, both models give the same result anyway.
"Fable 5 is Anthropic's step toward bringing Mythos-class capability to everyone — but with guardrails, a higher price, and 30-day data retention. The organizations that win are the ones that pick the right model for each task and set a clear data policy before they start."
- Saeree ERP Team
If your organization wants an ERP that keeps you in control of your data and is ready for AI ahead — contact the Saeree ERP team for a consultation.
