AA-PersonalityBench

Communication traits and measured writing characteristics describe how models express an answer.26 Oct 2026

One row per model and compatible score measure. Mean = equal-weight arithmetic average of imported reports. Distinct configurations reported for the same model in a source table remain separate. Reports can overlap; the range is not a confidence interval. Unbundle to inspect each observation.
		Evidence	What the result shows
Claude Fable 5.1 →Anthropic	100 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Claude Fable 5.1 →Anthropic	100 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Muse Wonder →Meta	100 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
GLM-5.3 →Z.ai	97.53 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Claude Fable 5.1 →Anthropic	97.46 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Claude Fable 5.1 →Anthropic	97.46 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Claude Fable 5.1 →Anthropic	97.46 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Claude Opus 5.5 →Anthropic	94.92 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
DeepSeek V4.1 Flash 1017 →DeepSeek	94.92 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
GPT-6 Astra →OpenAI	94.92 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Muse Wonder →Meta	94.92 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
GLM-5.3 →Z.ai	92.58 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Claude Opus 5.5 →Anthropic	92.38 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
GPT-6.1 Sol →OpenAI	92.38 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
GLM-5.3 →Z.ai	90.1 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
GLM-5.3 →Z.ai	90.1 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Claude Opus 5.5 →Anthropic	89.85 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Claude Opus 5.5 →Anthropic	89.85 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Muse Wonder →Meta	89.85 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Muse Wonder →Meta	89.85 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Kimi K3.1 Pro →Kimi	87.63 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Kimi K3.1 Pro →Kimi	87.63 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
DeepSeek V4.1 Flash 1017 →DeepSeek	87.31 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Muse Wonder →Meta	87.31 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
DeepSeek V4.1 Flash 1017 →DeepSeek	84.77 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Grok 4.7 →xAI	84.77 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Claude Opus 5.5 →Anthropic	82.23 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Gemini 4 Argon →Google	82.23 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
GLM-5.3 →Z.ai	80.2 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Claude Fable 5.1 →Anthropic	79.69 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Gemini 4 Argon →Google	79.69 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
GPT-6.1 Sol →OpenAI	79.69 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Claude Opus 5.5 →Anthropic	77.15 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Muse Wonder →Meta	77.15 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Gemini 4 Argon →Google	74.62 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
GPT-6 Astra →OpenAI	72.08 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
GPT-6 Astra →OpenAI	72.08 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Grok 4.7 →xAI	72.08 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Claude Opus 5.5 →Anthropic	69.54 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Gemini 4 Argon →Google	69.54 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
MiMo-V2.6-Pro →Xiaomi	69.54 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
DeepSeek V4.1 Flash 1017 →DeepSeek	67 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
GPT-6 Astra →OpenAI	67 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Grok 4.7 →xAI	64.46 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Kimi K3.1 Pro →Kimi	62.88 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Kimi K3.1 Pro →Kimi	62.88 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
GPT-6.1 Sol →OpenAI	61.92 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Gemini 4 Argon →Google	59.38 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
MiMo-V2.6-Pro →Xiaomi	59.38 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.
Muse Wonder →Meta	56.85 percentile 1 report	AA-PersonalityBench[1]	Descriptive measurement; read the named scale and configuration alongside the value.

Sources

AA-PersonalityBench: AI Personality & Communication Style Benchmark | Artificial Analysis ↗
Artificial Analysis
Dates & evaluation details
AA-PersonalityBench · expressiveness · 2026-10-26
Publisher-provided trait percentile, not a capability or correctness score. Higher means more of the named writing trait; desirability depends on the use case. Percentiles depend on the publisher’s full comparison population. This is a dated snapshot.
AA-PersonalityBench · organization · 2026-10-26
Publisher-provided trait percentile, not a capability or correctness score. Higher means more of the named writing trait; desirability depends on the use case. Percentiles depend on the publisher’s full comparison population. This is a dated snapshot.
AA-PersonalityBench · informativeness · 2026-10-26
Publisher-provided trait percentile, not a capability or correctness score. Higher means more of the named writing trait; desirability depends on the use case. Percentiles depend on the publisher’s full comparison population. This is a dated snapshot.
AA-PersonalityBench · transparency · 2026-10-26
Publisher-provided trait percentile, not a capability or correctness score. Higher means more of the named writing trait; desirability depends on the use case. Percentiles depend on the publisher’s full comparison population. This is a dated snapshot.
AA-PersonalityBench · warmth · 2026-10-26
Publisher-provided trait percentile, not a capability or correctness score. Higher means more of the named writing trait; desirability depends on the use case. Percentiles depend on the publisher’s full comparison population. This is a dated snapshot.
AA-PersonalityBench · conviction · 2026-10-26
Publisher-provided trait percentile, not a capability or correctness score. Higher means more of the named writing trait; desirability depends on the use case. Percentiles depend on the publisher’s full comparison population. This is a dated snapshot.
AA-PersonalityBench · imagination · 2026-10-26
Publisher-provided trait percentile, not a capability or correctness score. Higher means more of the named writing trait; desirability depends on the use case. Percentiles depend on the publisher’s full comparison population. This is a dated snapshot.
AA-PersonalityBench · slop · 2026-10-26
Publisher-provided trait percentile, not a capability or correctness score. Higher means more of the named writing trait; desirability depends on the use case. Percentiles depend on the publisher’s full comparison population. This is a dated snapshot.

TRACE THE SCORE · Per-model notes use published breakdowns where available.

What it measures

Communication traits and measured writing characteristics describe how models express an answer.

How to read the result

Style can matter for product fit. Compare expressiveness, organization, and transparency independently of factual capability.

Limits and comparison traps

A high trait percentile is a relative style measurement. Some traits are preferences rather than universally desirable abilities.

Benchmark reference notes

Measures communication style across 30 judged traits rolled into 8 indexes: Expressiveness, Organization, Transparency, Slop, Informativeness, Warmth, Conviction, and Imagination.

Where this fits in an evaluation suite

Format compliance, preference, truthfulness, and safety require separate measurements. Keep desirable writing style from masking factual errors.

Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.

A reproducible comparison

Use the exact implementation, version, task subset, tool access, attempt budget, and judge described by the source.

Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.

PRIMARY EVIDENCE & FURTHER READING

AA-PersonalityBench: AI Personality & Communication Style Benchmark | Artificial Analysis ↗
Artificial Analysis

Reviewed 2026-10-26. Explanation and interpretation by RLTasks. Benchmark ownership remains with its original authors.

Continue exploring

All instructions & behavior benchmarks →

Sources

What it measures

How to read the result

Limits and comparison traps

Benchmark reference notes

Where this fits in an evaluation suite

A reproducible comparison

Continue exploring

IFBench

IFEval

WildBench

HarmBench