Atlas

Models

← All models

Claude Fable 5.1

Anthropic · 2026-09-01

Claude Fable 5.1 and Claude Mythos 5.1 are deployments of Anthropic's September 2026 checkpoint with identical model weights. The default Fable 5.1 deployment is generally available with production safeguards; Mythos 5.1 relaxes domain-specific safeguards through trusted access programs. Safeguard interventions, fallback behavior, and API or consumer system prompts can change observed results and depend on the evaluation configuration.

Benchmark scores

BenchmarkScore
AA-3D v1.0 Elo1135
AA-AnalystAgent (pass@1)65.8
AA-AnalystAgent (pass@5)72.5
AA-AnalystAgent (pass^5)57.5
AA-Briefcase Elo1662
AA-Briefcase v1.1 Elo1669
AA-LCR v1.187.5
AA-Robotics v0.146.3
ActiveVision15.3
ActiveVision with tools76.5
AdvancedIF84.2
AdvancedIF (rubric mean score)95.4
Agent Red Teaming k=100 (CAIS)34.2
AIME 2025100.0
ALE-Bench2849
Android Bench 2.022.7
Android Bench 2.0 Completion82.4
Antidote: Everyday Edition1065
APEX (Mercor)26.7
APEX-Accounting61.0
APEX-Accounting Pass@111.7
APEX-Agents 1.177.1
APEX-Agents 1.1 Pass@168.6
APEX-SWE63.6
APEX-v1-extended73.8
ARC-AGI-1 Semi-Private97.5
ARC-AGI-290.0
Arena Alignment — Deceptive Completion10.5
Arena Alignment — False Attribution5.8
Arena Alignment — Unauthorized Action1.3
Loading Atlas data…