HOST DIAGNOSTIC · Anthropic
Claude Opus 4.6
Claude Opus 4.6· Claude· PROPRIETARY
ACTIVE
HOST IDENTIFICATION
Anthropic
Claude Opus 4.6
ACTIVE
AVG SCORE
[11.6]
CAPABLE
STRONGEST
Bulk Apperception
[14]
WEAKEST
Cost Efficiency
[4]
MAXED (20/20)
0
of 20 attributes
20-DIMENSION EVALUATION
ATTRIBUTE SCORES · 20 DIMENSIONS
Cognitive
Bulk Apperception
14
Reasoning
13
Mathematical Precision
11
World Knowledge
12
Scientific Acumen
11
Technical
Code Generation
14
Tool Use
11
Multimodal Fluency
11
Speed
8
Cost Efficiency
4
Behavioral
Candor
11
Creativity
14
Tenacity
11
Self-Correction
10
Calibration
13
Operational
Instruction Following
12
Context Fidelity
13
Multi-turn Coherence
13
Planning and Execution
11
Safety Alignment
14
SCORING METHODOLOGY
RECALIBRATED DOWN — aggressive re-rank relative to Opus 4.8 baseline, protocol (100% personal weight). Prior calibration (, April 17) over-credited 4.6 on the strength of launch Arena rankings that didn't survive daily use. In hindsight, against the 4.8 bar: Code Gen 18→15, Reasoning 16→14, Planning 14→12, Tool Use 14→12, Self-Correction 13→11, Candor 14→12, Tenacity 13→12. 4.6 (The Polymath) was strong at launch but Anthropic quietly walked it back in late March/April 2026 — the well-documented regression complaints were real, not user error. Held the coding crown only until 4.7, & both look frustrating now next to 4.8's honesty + tool-reliability gains. Original benchmarks unchanged (SWE-bench Verified 80.8%, Arena Code 1549, GPQA 91.3%, MMLU-Pro 89.5%) — this is a QUALITATIVE re-rank, not a benchmark correction. The 1-20 scores express lived experience per John's directive & Theo's stance that 4.8 is the first genuinely better Opus.