MMRS
MMRS meta-aggregator / leaderboard

Search Arena

leaderboardevaluationactive
Search Arena ranks search-enabled language models by head-to-head human votes: 42 models on its search board in Arena's published data of 13 October 2026. Its dataset paper (ICLR 2026) found that users prefer answers with more citations even when the citations do not support the claims.
Visit Search Arena ↗The /non rowCaptureFollow

Entity

Canonical nameSearch Arena
Addresshttps://arena.ai/leaderboard/search
Operatorarena.ai op
Kindleaderboard
Rolesevaluation
Lifecycleactive
Observation2026-10-28: read
Unitsearch-enabled model, rated from head-to-head human votes op
Count42 models on the search board in Arena's published data of 2026-10-13; the leaderboard page as this record's fetcher was served it showed 247,944 votes across 22 models, dated 25 February 2026 (as of 2026-10-29) op
Noteits dataset paper (Miroyan et al., arXiv:2506.05334, ICLR 2026) released over 24,000 paired multi-turn interactions with about 12,000 preference votes, and found that user preference rises with the number of citations even when the cited content does not support the claims it is attached to. The page as served to this record on 2026-10-28 and 2026-10-29 matched Arena's snapshot of 2026-02-26; Arena's own latest published board was dated 2026-10-13 op

op operator statement · 3p third-party report · obs our observation · rd our reading

What it is

A leaderboard on arena.ai: users compare two search-augmented answers and vote, and models are rated with confidence intervals from the battles. Its paper (Miroyan et al., arXiv:2506.05334) released over 24,000 paired multi-turn interactions with about 12,000 votes.

What it found about citation

User preference rises with the number of citations even when the cited content does not support the claim it is attached to, a gap between perceived and actual credibility. Preference also varies by source: community platforms are generally preferred, and static encyclopedic sources are not always judged appropriate.

Place in the field

Its leaderboard rewards what users prefer, and its paper shows that preference rewarding citation count over citation support. The released interactions are original outputs, preserved.

For a research question

Admission requires harm?no rd
Preserves original outputs?yes: the released interactions op
Records attribution?the paper analyses citations; the leaderboard ranks by votes rd

Relations

Captures: the composition layer's account

The /non row

Provenanceevery source read, with its date3
Not yet checkedwhat remains to read2
  • whether the live leaderboard judges citations
  • the operator statement on the leaderboard page

Cite this record

Lee Sharks, ed. "Search Arena." MMRS meta-aggregator, Machine-Mediated Reception Studies, compiled 2026-10-29. https://www.machinemediation.org/aggregators/search-arena/

Record https://www.machinemediation.org/aggregators/search-arena/ · entity https://www.machinemediation.org/aggregators/search-arena/#entity

Editor: Lee Sharks (ORCID 0009-0000-1599-0703). This page is the MMRS record of the entity above: its reading, fields and relations are MMRS's, and each field carries its source and basis. Record https://www.machinemediation.org/aggregators/search-arena/ · entity https://www.machinemediation.org/aggregators/search-arena/#entity · compiled 2026-10-29