name = "Claude Opus 4.8"
description = "Top Claude Opus tier for the hardest reasoning, coding, and long-horizon agents"
family = "claude-opus"
release_date = "2026-05-28"
last_updated = "2026-05-28"
attachment = true
reasoning = true
temperature = false
tool_call = true
open_weights = false
knowledge = "2026-01"

[limit]
context = 1_000_000
output = 128_000

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

[[benchmarks]]
name = "SWE-Bench Pro"
score = 69.2
metric = "resolve rate"
source = "https://www.anthropic.com/news/claude-opus-4-8"
date = "2026-05-28"

[[benchmarks]]
name = "Terminal-Bench"
score = 74.6
metric = "success rate"
harness = "Terminus-2"
version = "2.1"
source = "https://www.anthropic.com/news/claude-opus-4-8"
date = "2026-05-28"

[[benchmarks]]
name = "SWE-Bench Verified"
score = 88.6
metric = "resolved"
source = "https://benchlm.ai/benchmarks/sweVerified"

[[benchmarks]]
name = "Humanity's Last Exam"
score = 49.8
metric = "accuracy"
variant = "no tools"
source = "https://www.anthropic.com/news/claude-fable-5-mythos-5"
date = "2026-06-09"

[[benchmarks]]
name = "Humanity's Last Exam"
score = 57.9
metric = "accuracy"
variant = "with tools"
source = "https://www.anthropic.com/news/claude-fable-5-mythos-5"
date = "2026-06-09"

[[benchmarks]]
name = "OSWorld-Verified"
score = 83.4
metric = "success rate"
source = "https://www.anthropic.com/news/claude-fable-5-mythos-5"
date = "2026-06-09"

[[benchmarks]]
name = "FrontierCode"
score = 13.4
metric = "pass rate"
variant = "high effort"
dataset = "Diamond"
source = "https://www.anthropic.com/news/claude-fable-5-mythos-5"
date = "2026-06-09"