Atlas

Models

Anthropic

Claude Fable 5.2

170.6
Capability · 95% interval 168.5–173.1
Rank 1 of 893 (interval 1–2)
Relative cost 7.4× · $0.96 per short-answer task · $3.98 per agentic task

Claude Fable 5.2 and Claude Mythos 5.2 are deployments of the same underlying Anthropic model, released November 9, 2026. Fable 5.2 is generally available with cybersecurity and biology safeguards that answer blocked requests with Opus 4.8 or Opus 5; Mythos 5.2 relaxes those safeguards for organizations in Anthropic's Cyber and Life Sciences Verification Programs. Adaptive thinking is always on, with low, medium, high, xhigh, and max effort; high is the API default and medium the default in Claude apps and Claude Code. Safeguard interventions, fallback behavior, and API or consumer system prompts can change observed results and depend on the evaluation configuration. The published knowledge cutoff is June 2026, without day precision.

Launch post Model card

Scored on 209 benchmarks · 368 total scores · 332 independently reported

atlas.village.dev

Score providers

Organizations reporting headline scores for 209 benchmarks

In fitin the capability model (56 of 209)
Legal · 57 benchmarks · first place on 2
Immigration Bench - All-Passhigh
26.7 / 100Vals AI Immigration leaderboard
1 / 24———
GDPx: Legal
0.0 / 100GDPx leaderboard (updated 2026-11-09)
1 / 19——2026-11-09
HLAB-Vals-1.1 - Emerging Companies Venture Capitalhigh
30.0 / 100Vals AI HLAB-Vals-1.1 leaderboard
2 / 36———
HLAB-Vals-1.1 - White Collar Defense Investigationshigh
0.0 / 100Vals AI HLAB-Vals-1.1 leaderboard
2 / 36———
Immigration BenchIn fithigh
74.0 / 100Vals AI Immigration leaderboard
2 / 24~2 mo~Jan 2029—
Finance & Accounting · 42 benchmarks · first place on 6
Excel Modeling Benchmark v1.1In fithigh
77.4 / 100Vals AI EMB v1.1 leaderboard
1 / 40~2 mo~Oct 2027—
Excel Modeling Benchmark v1.1 Scratchhigh
73.6 / 100Vals AI EMB v1.1 leaderboard
1 / 40———
Excel Modeling Benchmark v1.1 LBOhigh
73.2 / 100Vals AI EMB v1.1 leaderboard
1 / 40———
Finance Agent v2.1In fithigh
69.5 / 100Vals AI Finance Agent v2.1 leaderboard
1 / 36~3 mo2028–2030—
Finance Agent v2.1 - Market Analysishigh
74.3 / 100Vals AI Finance Agent v2.1 leaderboard
1 / 36———
Code · 32 benchmarks · first place on 16
Vibe Code Bench v1.1In fithigh
93.8 / 100Vals AI Vibe Code Bench v1.1 leaderboard
1 / 117~3 moSep 2026—
SWE-Bench Pro (Public Dataset)In fit
93.0 / 100Claude Fable 5.2 and Claude Mythos 5.2 system card
1 / 76~3 mo~Dec 20262026-11-09
SWE-bench Multimodal
74.1 / 100Claude Fable 5.2 and Claude Mythos 5.2 system card
1 / 8——2026-11-09
SWE-bench Multimodal (Vals v1): Lighthousehigh
74.1 / 100Vals AI SWE-bench Multimodal leaderboard
1 / 20———
SWE-bench Multimodal (Vals v1): Scratch GUIhigh
50.0 / 100Vals AI SWE-bench Multimodal leaderboard
1 / 20———
Games · 16 benchmarks · first place on 5
Mystery Game PuzzlesIn fit
90.0 / 100Epoch AI benchmark hub (snapshot 2026-11-09)
1 / 80~4 mo~Mar 2027—
Chess Puzzles (task version 2.0)In fit
79.0 / 100Epoch AI benchmark hub (snapshot 2026-11-09)
1 / 88~4 mo~Jul 2027—
ChessBench — Mate in 1high
100.0 / 100ChessBench official run (2026-11-09T21-53-02-897Z-3b8bb58f)
1 / 15——2026-11-09
ChessBench — Hanging Piecehigh
66.7 / 100ChessBench official run (2026-11-09T21-53-02-897Z-3b8bb58f)
1 / 15——2026-11-09
EBR-bench v4 (card ban, single agent)In fit
90.5 / 100Epoch AI benchmark hub (snapshot 2026-11-09)
1 / 13~2 mo~Dec 2026—
General QA · 11 benchmarks · first place on 8
Humanity's Last ExamIn fit
73.2 / 100Claude Fable 5.2 and Claude Mythos 5.2 system card
1 / 142~2 mo~Sep 20272026-11-09
Humanity's Last Exam (with tools)In fit
74.2 / 100Claude Fable 5.2 and Claude Mythos 5.2 system card
1 / 77~3 mo~Jan 20282026-11-09
Humanity's Last Exam (Text-Only)
74.3 / 100Humanity's Last Exam (Text-Only) leaderboard (Artificial Analysis snapshot 2026-11-09)
1 / 489———
Humanity's Last Exam Diamond
80.2 / 100Humanity's Last Exam Diamond leaderboard (Artificial Analysis snapshot 2026-11-09)
1 / 139———
SimpleQA Verified (Epoch anti-abstention)In fit
94.0 / 100Epoch AI benchmark hub (snapshot 2026-11-09)
1 / 79~6 mo~May 2027—
Professional Work · 11 benchmarks · first place on 4
Remote Labor Index (RLI)In fitnot reported
41.3 / 100Scale Labs Remote Labor Index (RLI) leaderboard (snapshot 2026-11-09)
1 / 17~1 mo~Dec 2027—
GDPxIn fit
16.4 / 100GDPx leaderboard (updated 2026-11-09)
1 / 19~1 mo~Oct 20282026-11-09
GDPx: Chip Design
15.6 / 100GDPx leaderboard (updated 2026-11-09)
1 / 19——2026-11-09
GDPx: Hardware Design
21.0 / 100GDPx leaderboard (updated 2026-11-09)
1 / 19——2026-11-09
OfficeQA ProIn fit
72.9 / 100Claude Fable 5.2 and Claude Mythos 5.2 system card
2 / 13~6 mo~Sep 20272026-11-09
Multimodal · 9 benchmarks · first place on 6
HardGeoBench <=10minot reported
48.8 / 100HardGeoBench official leaderboard (snapshot 2026-11-09)
1 / 73———
HardGeoBench Country AccuracyIn fitnot reported
70.0 / 100HardGeoBench official leaderboard (snapshot 2026-11-09)
1 / 73~1.7 years2027–2031+—
HardGeoBench Avg Distancenot reported
552HardGeoBench official leaderboard (snapshot 2026-11-09)
1 / 73———
ChartographyIn fit
73.4 / 100Claude Fable 5.2 and Claude Mythos 5.2 system card
1 / 45~2 mo~May 20272026-11-09
Furniture Assembly BenchmarkIn fit
90.0 / 100Epoch AI benchmark hub (snapshot 2026-11-09)
1 / 38~2 mo~Dec 2026—
Science · 8 benchmarks · first place on 7
CritPt (avg@5)In fitx-high
35.7 / 100CritPt (avg@5) leaderboard (Artificial Analysis snapshot 2026-11-09)
1 / 403~1 mo~Mar 2028—
CritPt v2 (agentic, avg@5)In fit
86.3 / 100CritPt v2 (agentic, avg@5) leaderboard (Artificial Analysis snapshot 2026-11-09)
1 / 69~2 mo~Feb 2027—
BioMysteryBench (Vals Terminus 2)In fithigh
83.7 / 100Vals AI BioMysteryBench leaderboard
1 / 29~2 mo~Jun 2027—
BioMysteryBench — Human-solvable (Vals)high
92.2 / 100Vals AI BioMysteryBench leaderboard
1 / 29———
BioMysteryBench — Hard (Vals)high
47.1 / 100Vals AI BioMysteryBench leaderboard
1 / 29———
Agents · 7 benchmarks · first place on 4
OSWorld 2.1 Offline, 500-step Binary Accuracy
54.6 / 100Claude Fable 5.2 and Claude Mythos 5.2 system card
1 / 2——2026-11-09
OSWorld 2.1 Offline, 500-step Partial Score
89.3 / 100Claude Fable 5.2 and Claude Mythos 5.2 system card
1 / 4——2026-11-09
OSWorld 2.1 offline - reported partial reward, step limit unspecified
89.3 / 100Claude Fable 5.2 and Mythos 5.2 launch page
1 / 7——2026-11-09
AutomationBench - release-reported metric and version unspecified
51.4 / 100Claude Fable 5.2 and Mythos 5.2 launch page
1 / 5——2026-11-09
AutomationBench-AA 1.0.6In fitx-high
74.2 / 100AutomationBench-AA 1.0.6 leaderboard (Artificial Analysis snapshot 2026-11-09)
3 / 159~2 mo~May 2027—
Health & Medicine · 5 benchmarks · first place on 2
HealthBench Professional (Length-Adjusted)x-high
68.1 / 100HealthBench Professional (Length-Adjusted) leaderboard (Artificial Analysis snapshot 2026-11-09)
1 / 58———
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)In fit
43.9 / 100Medical Long Context Reasoning (MLCR-AA, October 2026 methodology) leaderboard (Artificial Analysis snapshot 2026-11-09)
1 / 108~4 mo~Apr 2029—
HealthBench ProfessionalIn fitx-high
71.4 / 100HealthBench Professional leaderboard (Artificial Analysis snapshot 2026-11-09)
2 / 38~3 mo~Jul 2028—
HealthBench (length-adjusted, Artificial Analysis)low
60.4 / 100Claude Fable 5.2 and Claude Mythos 5.2 system card
3 / 15——2026-11-09
HealthBenchIn fitlow
60.1 / 100Claude Fable 5.2 and Claude Mythos 5.2 system card
15 / 45~1.0 yearsafter 20312026-11-09
Indexes · 4 benchmarks · first place on 3
Epoch Capabilities Index (ECI)not reported
174Epoch AI benchmark hub (snapshot 2026-11-09)
1 / 253———
Anthropic ECI (AECI), November 2026 fit
177Claude Fable 5.2 and Claude Mythos 5.2 system card
1 / 6——2026-11-09
Capability
170.6 / 171Capability
1 / 893——2026-11-10
Vals Finance Indexhigh
63.6 / 100Vals AI Vals Finance Index leaderboard
4 / 36———
Math · 4 benchmarks · first place on 3
FrontierMath Tier 4 (v2) - PrivateIn fit
100.0 / 100Epoch AI benchmark hub (snapshot 2026-11-09)
1 / 72~3 moFeb 2026—
FrontierMath Tiers 1-3 (v2) - PrivateIn fit
95.1 / 100Epoch AI benchmark hub (snapshot 2026-11-09)
1 / 89~3 moSep 2026—
ProofBench v1.1In fithigh
100.0 / 100Vals AI ProofBench v1.1 leaderboard
1 / 51~3 moFeb 2026—
OTIS Mock AIME 2024-2025In fit
99.7 / 100Epoch AI benchmark hub (snapshot 2026-11-09)
15 / 139~1.0 yearsApr 2025—
Cybersecurity · 1 benchmark
GDPx: Cybersecurity
9.3 / 100GDPx leaderboard (updated 2026-11-09)
9 / 19——2026-11-09
Safety · 1 benchmark · first place on 1
GDPx: Alignment
9.3 / 100GDPx leaderboard (updated 2026-11-09)
1 / 19——2026-11-09
Search · 1 benchmark
AA-LCR v1.1In fithigh
83.3 / 100AA-LCR v1.1 leaderboard (Artificial Analysis snapshot 2026-11-09)
17 / 407~6 mo~Apr 2027—
Loading Atlas data…