# Launch evaluations enable production safeguards: cyber tasks may fall back to Opus 4.8, biology tasks to Opus 5.
# OSWorld uses the August 2026 task release and scores safeguard interventions as zero, without fallback.
name = "Claude Fable 5.1"
description = "Claude model for demanding reasoning and long-horizon agentic work"
family = "claude-fable"
release_date = "2026-09-01"
last_updated = "2026-09-01"
attachment = true
reasoning = true
temperature = false
tool_call = true
open_weights = false
knowledge = "2026-06"

[limit]
context = 1_000_000
output = 128_000

[modalities]
input = ["text", "image", "pdf"]
output = ["text"]

[[benchmarks]]
name = "Terminal-Bench-Science"
score = 52.6
metric = "score"
variant = "production safeguards with fallback"
version = "0.1"
source = "https://www.anthropic.com/claude-fable-and-mythos-5-1"
date = "2026-09-01"

[[benchmarks]]
name = "Terminal-Bench"
score = 55.8
metric = "score"
variant = "production safeguards with fallback"
version = "4.0"
source = "https://www.anthropic.com/claude-fable-and-mythos-5-1"
date = "2026-09-01"

[[benchmarks]]
name = "GDPval-AA"
score = 1853
metric = "Elo"
variant = "production safeguards with fallback"
version = "2"
source = "https://www.anthropic.com/claude-fable-and-mythos-5-1"
date = "2026-09-01"

[[benchmarks]]
name = "OSWorld"
score = 77.9
metric = "partial score"
variant = "production safeguards; interventions score zero; no fallback"
version = "2.0"
source = "https://www.anthropic.com/claude-fable-and-mythos-5-1"
date = "2026-09-01"
dataset = "August 2026 task release"

[[benchmarks]]
name = "OSWorld"
score = 41.7
metric = "strict success rate"
variant = "production safeguards; interventions score zero; no fallback"
version = "2.0"
source = "https://www.anthropic.com/claude-fable-and-mythos-5-1"
date = "2026-09-01"
dataset = "August 2026 task release"

[[benchmarks]]
name = "Humanity's Last Exam"
score = 60.9
metric = "score"
variant = "without tools; production safeguards with fallback"
source = "https://www.anthropic.com/claude-fable-and-mythos-5-1"
date = "2026-09-01"

[[benchmarks]]
name = "Humanity's Last Exam"
score = 65
metric = "score"
variant = "with tools; production safeguards with fallback"
source = "https://www.anthropic.com/claude-fable-and-mythos-5-1"
date = "2026-09-01"

[[benchmarks]]
name = "AutomationBench"
score = 31.4
metric = "pass rate"
variant = "production safeguards with fallback"
source = "https://www.anthropic.com/claude-fable-and-mythos-5-1"
date = "2026-09-01"

[[benchmarks]]
name = "CursorBench"
score = 73.4
metric = "score"
variant = "max effort; production safeguards with fallback"
version = "3.2.0"
source = "https://www.anthropic.com/claude-fable-and-mythos-5-1"
date = "2026-09-01"