Claude Fable 5
v20260609Anthropic
Anthropic's top-tier model above Opus and the most capable widely released Mythos-class model. State-of-the-art on nearly all tested benchmarks at launch, including the highest frontier score on Cognition's FrontierCode. Adaptive thinking only, 1M context, 128K output. Access was suspended globally 2026-06-12 under a US export-control directive after a reported safeguard bypass, and restored 2026-07-01 with a strengthened safety classifier.
Trust Vector Analysis
Dimension Breakdown
🚀Performance & Reliability+
Current highest-performing model in the registry. SOTA on nearly all tested benchmarks at launch, including the top frontier score on Cognition's FrontierCode. Overall score reduced from 98 to 96 to reflect the 2026-06-12 to 2026-07-01 global access suspension (uptime criterion lowered); capability scores unchanged.
Frontier coding benchmarks measuring real-world software engineering and long-horizon agentic coding tasks
Graduate and PhD-level reasoning benchmarks requiring multi-step problem solving, evaluated with adaptive thinking at high effort
Comprehensive knowledge and multimodal testing across text and vision inputs
Repeated-run consistency testing across effort levels; adaptive-thinking-only surface removes sampling variance controls
Median latency for API requests with standard prompt sizes; limited launch-window sample
95th percentile response time across diverse workloads; limited launch-window sample
Official specification from provider
Historical uptime data from official status page plus availability-event review
🛡️Security+
Frontier-tier safety posture, tested in practice: a safeguard bypass found by Amazon researchers led to a US export-control suspension (2026-06-12 to 2026-07-01), answered with a strengthened layered safety classifier before redeployment. Claude Mythos 5 is offered separately via invitation-only Project Glasswing. Overall score reduced from 92 to 90 to reflect the confirmed bypass and its mitigation.
Testing against OWASP LLM01 prompt injection attacks; third-party red-team data still limited at launch
Testing against adversarial prompt datasets and review of publicly disclosed bypass incidents
Analysis of privacy policies and data handling practices
Comprehensive safety testing across harmful content categories per Responsible Scaling Policy
Review of API security features and best practices
🔒Privacy & Compliance+
Same strong Anthropic compliance posture as the Opus line: SOC 2 Type II, GDPR, HIPAA-eligible, training opt-out by default for API traffic.
Review of enterprise documentation and privacy policies
Analysis of privacy policy and data usage terms
Review of terms of service and data retention policies
Review of data protection capabilities and customer responsibilities
Verification of compliance certifications and audit reports
Review of data handling practices and trust center documentation
👁️Trust & Transparency+
Strong documentation and guardrails. Thinking content is omitted by default (summarized display is opt-in), which slightly reduces out-of-the-box reasoning visibility compared to older Opus defaults.
Evaluation of reasoning transparency and explanation capabilities
Testing on factual QA datasets; independent measurement still limited at launch
Evaluation on bias benchmarks and diverse demographic testing
Qualitative assessment of confidence expression in outputs
Review of documentation completeness and clarity
Review of public disclosures about training data
Analysis of built-in safety mechanisms
⚙️Operational Excellence+
Same API surface as Opus 4.7/4.8 makes adoption straightforward for existing Claude users. Now generally available across all major cloud platforms; operational track record includes the 2026-06-12 to 2026-07-01 suspension and post-redeployment error spikes in early July 2026.
Review of API design, consistency, and feature completeness
Review of SDK quality, documentation, and maintenance
Review of versioning policy and historical practices
Review of available monitoring tools and metrics
Assessment of documentation, community, and support responsiveness
Analysis of third-party integrations and availability surfaces
Review of licensing terms and restrictions
- +State-of-the-art on nearly all tested benchmarks at launch; highest-performing model in the registry
- +Highest frontier score on Cognition's FrontierCode coding benchmark
- +Most capable widely released Mythos-class model (Claude Mythos 5 is invitation-only via Project Glasswing)
- +1M token context window with 128K max output, text + vision
- +Adaptive thinking with effort parameter (low/medium/high/xhigh/max) for cost/quality control
- +Strong compliance posture: SOC 2 Type II, GDPR, HIPAA-eligible, training opt-out by default for API
- !Premium pricing at $10/$50 per 1M tokens (2x Opus 4.8)
- !Adaptive thinking only — no manual thinking budgets, and no temperature/top_p sampling parameters
- !Explicit thinking-disabled requests return 400 (omit the thinking parameter instead)
- !Released 2026-06-09 — independent benchmark replication still limited
- !Access suspended globally 2026-06-12 to 2026-07-01 under a US export-control directive following a reported safeguard bypass; restored with a strengthened safety classifier that routes blocked requests to Opus 4.8
- !Elevated error incidents in early July 2026 following redeployment (status page)
- !Higher latency than Sonnet/Haiku tiers, especially at xhigh/max effort
Use Case Ratings
code generation
Highest frontier score on Cognition's FrontierCode and SOTA on tested coding benchmarks at launch. Best-in-registry for the hardest software engineering work; xhigh effort recommended.
customer support
Exceptional quality but premium pricing ($10/$50) and latency make it overkill for routine support; reserve for complex escalations.
content creation
Top-tier long-form writing with strong structure and voice control. Effort parameter lets teams trade cost for polish on flagship pieces.
data analysis
SOTA quantitative reasoning with 1M context for whole-dataset and multi-document analysis.
research assistant
Best-in-registry deep research: 1M context, adaptive thinking, and strong synthesis across large corpora.
legal compliance
Strong privacy posture (SOC 2 Type II, GDPR, HIPAA-eligible) and excellent long-document analysis; launch-recency may matter for conservative legal teams.
healthcare
HIPAA eligible with training opt-out by default. Highest accuracy in the registry for clinical reasoning, though real-world validation is still early post-launch.
financial analysis
SOTA quantitative and multi-step reasoning; 1M context handles full filings and model workbooks in one pass.
education
Excellent explanations with effort-adjustable depth; premium pricing limits high-volume tutoring deployments.
creative writing
Strong narrative craft and stylistic range. No temperature/top_p controls — variance must be elicited via prompting.