HOST DIAGNOSTIC · Anthropic
Claude Opus 5
Claude Opus 5· Claude· PROPRIETARY
ACTIVE
HOST IDENTIFICATION
Anthropic
Claude Opus 5
ACTIVE
AVG SCORE
[15.8]
STRONG
STRONGEST
Code Generation
[19]
WEAKEST
Speed
[8]
MAXED (20/20)
0
of 20 attributes
20-DIMENSION EVALUATION
ATTRIBUTE SCORES · 20 DIMENSIONS
Cognitive
Bulk Apperception
16
Reasoning
17
Mathematical Precision
18
World Knowledge
15
Scientific Acumen
16
Technical
Code Generation
19
Tool Use
17
Multimodal Fluency
15
Speed
8
Cost Efficiency
10
Behavioral
Candor
17
Creativity
14
Tenacity
17
Self-Correction
17
Calibration
16
Operational
Instruction Following
17
Context Fidelity
16
Multi-turn Coherence
15
Planning and Execution
17
Safety Alignment
18
SCORING METHODOLOGY
PROVISIONAL CALIBRATION per /NM310-A1 (2026-09-07) — 0 BAND / 0 BAND-COMPOSITE / 19 INHERITED-PROVISIONAL / 1 PROXY; re-calibrate when per-dimension evals land. Anthropic Opus-tier flagship, released 2026-07-24. Positioned as near-Fable 5 frontier intelligence at half the price ($5/$25 per M in/out, same as Opus 4.8); 1M-token context (default + max), 128K output, thinking on by default, effort toggle (low/med/high) + Fast mode; new default on Claude Max, strongest on Claude Pro. Anthropic claims new SOTA on Frontier-Bench (43.3% per launch materials — NO Benchmark record exists yet, intake item) + GDPval-AA (no numeric Elo published in announcement, so no eval snapshot per NM108); remains behind Mythos 5 on cybersecurity tasks. ARC-AGI-3 30.2% captured as eval snapshot (self-reported at launch; re-check arcprize.org for verified scorecard). Framed as most-aligned Opus + strongest generally available model for scientific research (biology gains vs 4.8). Fourth Claude 5-family release in under two months (Mythos/Fable/Sonnet June 9, Opus July 24; Haiku still 4.x). PRIMARY SOURCE: https://www.anthropic.com/news/claude-opus-5. Corroboration: Bloomberg, TechCrunch, Axios, Fortune (2026-07-24). PROVENANCE (-A1 lanes M1/M1b, per-dim): Cost Efficiency [PROXY:AA $10.00/M blended]; all remaining 19 dimensions [INHERITED-PROVISIONAL:claude-opus-4-8]. ARC-AGI-3 30.2% is on file but banding it (6-8) was an 11-point downgrade vs the predecessor and was not confirmed, so Planning and Execution stays inherited (17). Header normalised to the Amendment A1 form on 2026-09-07 (was 'CALIBRATED … 0 BAND / 19 INHERITED / 1 PROXY').