Anthropic

Claude Opus 4.8

Released May 28, 2026 · knowledge cutoff 2026-01

Status: Superseded
Location: United States
Modality: Multimodal
Context window: 1M
Max output: 128K
Speed: —
Price ($/MTok in / out): $5 / $25
Cost per task (max effort): $1.80
Open weights: No

Benchmarks5 of 8 reported

GPQA Diamond

93.6%

SWE-bench Verified

88.6%

Humanity's Last Exam

49.8%

LMArena Elo

1484

ARC-AGI-2

72.1%

MMLU-Pro

—

Terminal-Bench 2.1

—

AIME (latest)

—

Strengths & weaknesses

Strengths

Flags its own uncertainty and asks rather than guessing — far less confident nonsense
The effort setting genuinely changes token spend, so quick and deep work share one model
More expressive and varied on creative work than the 4.x releases before it

Weaknesses

The extra hedging is friction when you want a confident answer to a low-stakes question
Can get caught in self-correction loops, re-checking work that was already right
Clarifying questions interrupt fully autonomous runs that were meant to need no input

HLE 49.8% no tools / 57.9% with tools. LMArena Elo corrected to 1484 (current Opus 4.8 Thinking listing, July 2026); an earlier 1510 figure did not match the live leaderboard. ARC-AGI-2 72.1% added July 2026 once ARC Prize published verified results: 72.08% at high effort (71.67% medium, 62.22% low).

For developers

API model strings

Not researched

Licence

Proprietary — weights not released

Retirement

No retirement announced

Lineage

No recorded predecessor