Claude Fable 5: First Public Mythos-Class AI Model - Pricing, Benchmarks and Guardrails

Anthropic released Claude Fable 5 on June 9, 2026 - the first publicly available Mythos-class model. Priced at $10/$50 per million tokens, it delivers 80.3% on SWE-Bench Pro but reroutes cybersecurity, biology and chemistry queries to Opus 4.8.

Core Verdict

Anthropic released Claude Fable 5 on June 9, 2026 — the first publicly available Mythos-class model. Priced at $10/$50 per million tokens (2x Opus 4.8), Fable 5 delivers 80.3% on SWE-Bench Pro and 88% on Terminal-Bench 2.1, but ships with safety guardrails that reroute cybersecurity, biology, and chemistry queries to the older Opus 4.8. The unrestricted Claude Mythos 5 remains limited to vetted Project Glasswing partners.

What Happened

On June 9, 2026, Anthropic announced two new frontier models — Claude Fable 5 and Claude Mythos 5 — marking the company's first broad release of its previously restricted Mythos-class technology.

Detail	Value
Release Date	June 9, 2026
Class	Mythos (first publicly available)
Context Window	1M tokens (default)
Max Output	128K tokens per request
API Pricing	$10/1M input, $50/1M output
Plan Availability	Pro, Max, Team, Enterprise (free June 9-22, then usage credits)
Cloud Availability	Anthropic API, AWS Bedrock, GCP Vertex AI

The underlying model is shared between Fable 5 and Mythos 5. The difference: safeguards. Fable 5 uses conservative safety classifiers that reroute high-risk queries to Claude Opus 4.8, while Mythos 5 operates with fewer restrictions, available only to Anthropic's Project Glasswing partners and select biology researchers.

"Claude Fable 5 is the first to break 90% on our core analytics benchmark of complex, long-running analytical tasks — a 10-point jump over Opus," Anthropic wrote in its launch announcement.

Guardrails: What Fable 5 Blocks

The defining feature of Claude Fable 5 is not what it can do — it's what it won't do.

Blocked Domain	Behavior	Trigger Rate
Cybersecurity	Rerouted to Opus 4.8	<5% of sessions
Biology	Rerouted to Opus 4.8	<5% of sessions
Chemistry	Rerouted to Opus 4.8	<5% of sessions
Model Distillation	Rerouted to Opus 4.8	<5% of sessions
AAV Research	Blocked (capsid modification queries)	Rare

Anthropic's rationale: Mythos-class models pose "substantial risk of uplift to malicious actors" in frontier cybersecurity and research biology. The system card published alongside the release details extensive evaluations showing that Mythos-class models could significantly lower the barrier to offensive cyber operations and biological weapon design.

The safeguards operate as a two-tier system:

Overt blocking — Queries explicitly about cybersecurity exploits, biological weapons, or chemical synthesis are refused or rerouted to Opus 4.8. Users see a clear fallback notification.
Covert blocking — Certain high-risk capabilities (distillation attempts, AAV capsid modification) are silently blocked without user visibility. The model reroutes to a safe response without indicating the intervention.

The reroute is not a simple "I can't answer that." Fable 5 silently hands off to Opus 4.8, which may still answer the question at a lower capability level. This design choice has sparked controversy in the developer community — some argue it undermines trust in the model's transparency.

Benchmarks: Where Fable 5 Dominates

Anthropic published comprehensive benchmark data comparing Fable 5 / Mythos 5 against Claude Opus 4.8, GPT-5.5, and Gemini 3.1 Pro:

Fable 5 benchmark comparison chart

Benchmark	Fable 5 / Mythos 5	Opus 4.8	GPT-5.5	Gemini 3.1 Pro
SWE-Bench Pro	80.3%	~55%	58.6%	54.2%
SWE-Bench Verified	95%	—	—	—
Terminal-Bench 2.1	88.0%	82.7%	83.4%	70.7%
Analytics Benchmark (Hex)	90%+	~80%	—	—
AIME 2025 (Math)	~92%	~82%	~88%	~85%
MMLU-Pro	~89%	~84%	~86%	~83%

The standout result is SWE-Bench Pro at 80.3% — a massive 22-point lead over GPT-5.5 and 26 points over Gemini 3.1 Pro. This means Fable 5 can autonomously resolve 8 out of 10 real-world software engineering tasks, a capability that puts it firmly in "junior-to-mid-level developer" territory for autonomous coding.

On Terminal-Bench 2.1, which tests long-horizon terminal-based agentic tasks, Fable 5's 88% represents a 5.3-point improvement over Opus 4.8 — significant for users running complex automation pipelines.

Pricing Reality Check

At $10 per million input tokens and $50 per million output tokens, Claude Fable 5 is the most expensive frontier model on the market:

Model	Input $/1M	Output $/1M	Cost Ratio (vs Fable 5)
Claude Fable 5	$10	$50	1x (baseline)
Claude Opus 4.8	$5	$25	0.5x
Claude Sonnet 4.6	$3	$15	0.3x
GPT-5.5	$2	$10	0.2x
Gemini 3.5 Flash	~$0.35	~$1.40	~0.03x

The 2x price over Opus 4.8 is the headline, but the real concern is the 5x vs GPT-5.5 gap on output tokens. For heavy coding agent use (which is exactly where Fable 5 excels), this pricing premium adds up fast.

Mitigation factors:

Free access window: June 9-22 on Pro, Max, Team, and Enterprise plans (no extra cost)
Caching: Prompt caching at $1/MTok read, $12.50/MTok write (5-min), $20/MTok write (1-hour)
Batch API: 50% discount at $5/$25 per million tokens

Despite the high price, the benchmark lead on SWE-Bench Pro (80.3% vs 58.6% GPT-5.5) means that for tasks requiring autonomous resolution of complex engineering issues, Fable 5 may actually be cheaper per successfully resolved task — you pay more per token but complete the job in fewer attempts.

HN Community Reaction

The Hacker News thread on Fable 5's release has 2,159 points and 1,669 comments — one of the most active AI threads of 2026. Key themes:

Green: Enthusiasm for capabilities — Developers testing Fable 5 report dramatic improvements in long-horizon coding tasks. Several commenters share stories of Fable 5 resolving complex refactoring and debugging tasks that stymied Opus 4.8 and GPT-5.5.

Yellow: Concern about guardrails — The safety rerouting is the most debated feature. Critics argue the overt blocking of cybersecurity queries undermines the model's usefulness for security researchers. Anthropic's response: partnered cybersecurity professionals get Mythos 5 access through Project Glasswing.

Red: Pricing anxiety — The $50/MTok output rate generates significant pushback. "At this price, Fable 5 is for enterprise teams with dedicated AI budgets, not indie developers," writes one top comment.

Purple: The 'two-tier AI' debate — Several threads argue that Fable 5 vs Mythos 5 represents a worrying precedent — a powerful model reserved for the connected few, while the public gets a sanitized version. Anthropic counters that this is responsible deployment, not gatekeeping.

What This Means for You

For Developers and AI Tool Users

Code more accurately, not more cheaply: If you are using Claude Code or Cursor with Opus 4.8 today, upgrading to Fable 5 will resolve more complex bugs autonomously — but cost ~2x per session. For simple completion tasks, stick with Sonnet 4.6.
Free access window is crucial: The June 9-22 window on existing plans is your chance to evaluate Fable 5 at zero marginal cost. Run your hardest automation pipelines and long-running agentic tasks during this period to gauge whether the upgrade justifies the premium.
Know the guardrail boundaries: If your workflow touches cybersecurity (penetration testing, vulnerability research) or biology/chemistry analysis, Fable 5 will silently fall back to Opus 4.8. Plan for this — either use Mythos 5 (if you qualify) or keep a secondary pipeline with GPT-5.5 for those domains.

For Affiliate and Tool Strategy

This is a clear signal to deepen the Claude cluster on WayToClawEarn:

Update existing pricing guides with Fable 5's $10/$50 rate
Add a Fable 5 vs Opus 4.8 cost-comparison calculator
Create a Tutorial on setting up prompt caching to offset Fable 5 costs
The guardrail controversy is a strong angle for a Case article: "How security researchers work around AI model restrictions"