# Launch evaluations use adaptive thinking and production safeguards, with Opus 4.8 fallback for cyber and Opus 5 for bio/frontier-LLM development.
# AutomationBench disables fallback and counts safeguard-triggered tasks as failures; Terminal-Bench uses xhigh effort.
name = "Claude Opus 5.5"
description = "Claude model for long-running agentic coding and knowledge work"
family = "claude-opus"
release_date = "2026-09-22"
last_updated = "2026-09-22"
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"
score = 66.4
metric = "score"
variant = "xhigh effort; production safeguards with fallback"
version = "4.0"
source = "https://www.anthropic.com/claude-opus-5-5"
date = "2026-09-22"

[[benchmarks]]
name = "FrontierCode"
score = 54.4
metric = "score"
variant = "max effort; production safeguards with fallback"
version = "1.1"
dataset = "Main"
source = "https://www.anthropic.com/claude-opus-5-5"
date = "2026-09-22"

[[benchmarks]]
name = "CursorBench"
score = 57.8
metric = "score"
variant = "max effort; production safeguards with fallback"
version = "4.0"
source = "https://www.anthropic.com/claude-opus-5-5"
date = "2026-09-22"

[[benchmarks]]
name = "GDPval-AA"
score = 1846
metric = "Elo"
variant = "max effort; production safeguards with fallback"
version = "2.1"
source = "https://www.anthropic.com/claude-opus-5-5"
date = "2026-09-22"

[[benchmarks]]
name = "AutomationBench"
score = 40
metric = "pass rate"
variant = "max effort; production safeguards; no fallback"
source = "https://www.anthropic.com/claude-opus-5-5"
date = "2026-09-22"

[[benchmarks]]
name = "Humanity's Last Exam"
score = 67.7
metric = "score"
variant = "max effort; with tools; production safeguards with fallback"
source = "https://www.anthropic.com/claude-opus-5-5"
date = "2026-09-22"

[[benchmarks]]
name = "Terminal-Bench-Science"
score = 58.7
metric = "score"
variant = "max effort; production safeguards with fallback"
version = "0.1"
source = "https://www.anthropic.com/claude-opus-5-5"
date = "2026-09-22"

[[benchmarks]]
name = "OSWorld"
score = 81.8
metric = "partial score"
variant = "max effort; production safeguards with fallback"
version = "2.0"
source = "https://www.anthropic.com/claude-opus-5-5"
date = "2026-09-22"

[[benchmarks]]
name = "Chartography"
score = 89
metric = "score"
variant = "max effort; with tools; production safeguards with fallback"
source = "https://www.anthropic.com/claude-opus-5-5"
date = "2026-09-22"