Human preference leaderboard
Arena
What people prefer when two answers are put side by side.
Key information
- Focus
- Human preferences between model responses; Text Arena is the scope described here.
- Measure
- Text Arena rating, published uncertainty bounds and vote counts.
- Version / scope
- Text · overall · style-controlled
- Update pattern
- Published snapshots change as votes and eligible models are added; our excerpt is a dated manual edition, not a live feed.
- Publisher update
- 13 October 2026
- Editorial verification
- 15 October 2026
In plain English
What makes it different?
Arena asks people to compare anonymous responses to the same prompt. Those choices are evidence about which response people prefer in that setting. They do not independently establish factual correctness, safety or success on a specific technical task.
Sources and attribution are listed below.
What it measures
Human preferences between model responses; Text Arena is the scope described here.
Strengths
- Captures preferences on prompts supplied by actual users.
- Publishes uncertainty and a documented rating pipeline.
Limitations
- The voting population and its prompts need not represent your own users.
- Style, familiarity and which models meet each other can affect comparisons. Overlapping intervals do not establish a clear winner.
How the evaluation works
Pairwise votes feed a statistical rating model. Text ratings can apply style controls and sampling adjustments; changing the filter changes the comparison.
We show a small excerpt from the publisher's separately licensed numerical dataset. Its model labels, order and intervals are retained; no scor.es weighting is applied.
Configuration matters
Text, vision, search, coding and agent leaderboards are different evaluations. A model label or effort variant in this list is not automatically the same configuration as another benchmark or a consumer app.
Primary methods: Leaderboard policy ↗Arena-Rank methodology release ↗
Publisher's Top 5
Text Arena, style control · overall. Published 13 October 2026; checked 15 October 2026. A dated publisher snapshot, not a scor.es ranking or a live feed.
| Rank | Publisher model identifier | Rating | Published bounds |
|---|---|---|---|
| 1 | gemini-4-argon-highgoogle | 1,524.8Bounds 1,515.9–1,533.6 | 1,515.9–1,533.6 |
| 2 | muse-wonder-maxmeta | 1,512.7Bounds 1,499.4–1,525.9 | 1,499.4–1,525.9 |
| 3 | claude-opus-5.5-highanthropic | 1,508.1Bounds 1,500.5–1,515.6 | 1,500.5–1,515.6 |
| 4 | claude-fable-5-highanthropic | 1,503.4Bounds 1,499.1–1,507.7 | 1,499.1–1,507.7 |
| 5 | claude-opus-4-6-highanthropic | 1,503.1Bounds 1,499.8–1,506.5 | 1,499.8–1,506.5 |
Source: Arena, official numerical dataset ↗. CC BY 4.0. Selected ranks 1–5; display rounded to one decimal. Overlapping bounds do not establish a clear winner. No endorsement is implied.
Versions & freshness
Verified 15 October 2026The excerpt is the 13 October 2026 overall Text style-control snapshot. This is its publication date, not the date of every vote or a claim that no newer website view exists.
Official Text leaderboard and licensed dataset accessible. Published snapshots change as votes and eligible models are added; our excerpt is a dated manual edition, not a live feed.
Our review date is not an evaluation date. Individual run dates are not consistently available; they remain unknown here. No automatic result collection is enabled.
Publication & reuse
The official numerical dataset is CC BY 4.0. This permits reuse of that dataset with attribution, a licence link and notice of changes. It does not license conversations, model weights, logos or another publisher's data.
Arena-Rank's Apache 2.0 code licence is separate from the numerical dataset's CC BY 4.0 licence.
View the scope of the reuse decision
- Result fields
- model_name, organization, rank, rating, rating_lower, rating_upper, vote_count, category, leaderboard_publish_date
- Public result display
- conditional
- Storage
- conditional
- Historical copies
- conditional
- Updates
- conditional
- Commercial reuse
- conditional
- Derived ratings
- Only these licensed Arena inputs, subject to attribution and notice of changes. Does not clear mixtures with other publishers' inputs.
- Attribution
- Arena, lmarena-ai/leaderboard-dataset, CC BY 4.0. Link the pinned source and licence; identify selection and display rounding. No endorsement.
This decision does not clear mixed-source scientific recommendations. Publisher content, software, datasets and model licences are separate. No partnership is implied.
Primary sources & attribution
This guide is original scor.es commentary about Arena's work. The links below support the identity, methods, versions and reuse review.
- Official Text leaderboard ↗Checked 15 October 2026
- Leaderboard policy ↗Checked 15 October 2026
- Official numerical dataset (pinned edition) ↗Checked 15 October 2026
- CC BY 4.0 licence ↗Checked 15 October 2026
- Arena-Rank methodology release ↗Checked 15 October 2026
Editorial edition 2026-10-15.1 · reuse review directory-rights-1. Re-check due 14 November 2026.