Image-to-Video Arena: LMArena Image-to-Video Arena
Crowdsourced human-preference leaderboard for image-to-video models run by LMArena. Voters compare two anonymous clips generated from the same input image and prompt and pick the better one; ratings use a Bradley-Terry model with 95% confidence intervals. Higher is better.
Configurations
7
tracked on this benchmark
Score band
1488.0 – 1108.0
best → lowest tracked
Snapshot trend
—
need ≥2 snapshots
Leaderboard
Tracked models ranked by Arena Score (higher is better).
Configuration: Overall leaderboard entry (unqualified board entry)
Notes: Arena score 1488 (95% CI 1478-1499), 3,734 votes, rank 2 of 48, category overall, leaderboard published 2026-09-22. Board entry 'gemini-omni-1.1-flash'. Source: LMArena (CC BY 4.0). Retrieved 2026-10-17.
Configuration: With audio (Gemini API default for this model code: audio always on)
Notes: Arena score 1331 (95% CI 1320-1343), 32,367 votes, rank 20 of 48, category overall, leaderboard published 2026-09-22. Board entry 'veo-3-audio'. Audio entry chosen per Google's Gemini API docs (Veo 3 audio always on, https://ai.google.dev/gemini-api/docs/veo); no-audio entry 'veo-3' kept as alternate. Source: LMArena (CC BY 4.0). Retrieved 2026-10-17.
Configuration: With audio (Gemini API default for this model code: audio always on)
Notes: Arena score 1326 (95% CI 1317-1336), 41,194 votes, rank 21 of 48, category overall, leaderboard published 2026-09-22. Board entry 'veo-3-fast-audio'. Audio entry chosen per Google's Gemini API docs (Veo 3 Fast audio always on, https://ai.google.dev/gemini-api/docs/veo); no-audio entry 'veo-3-fast' kept as alternate. Source: LMArena (CC BY 4.0). Retrieved 2026-10-17.
Configuration: Overall leaderboard entry (unqualified board entry)
Notes: Arena score 1307 (95% CI 1300-1314), 245,206 votes, rank 24 of 48, category overall, leaderboard published 2026-09-22. Board entry 'seedance-v1.5-pro'. Source: LMArena (CC BY 4.0). Retrieved 2026-10-17.
Configuration: Overall leaderboard entry (unqualified board entry)
Notes: Arena score 1226 (95% CI 1207-1245), 1,585 votes, rank 37 of 48, category overall, leaderboard published 2026-09-22. Board entry 'ray-3'. Source: LMArena (CC BY 4.0). Retrieved 2026-10-17.
Configuration: Overall leaderboard entry (unqualified board entry)
Notes: Arena score 1166 (95% CI 1150-1182), 10,303 votes, rank 44 of 48, category overall, leaderboard published 2026-09-22. Board entry 'veo-2'. Source: LMArena (CC BY 4.0). Retrieved 2026-10-17.
Configuration: Overall leaderboard entry (unqualified board entry)
Notes: Arena score 1108 (95% CI 1092-1125), 9,521 votes, rank 46 of 48, category overall, leaderboard published 2026-09-22. Board entry 'ray2'. Source: LMArena (CC BY 4.0). Retrieved 2026-10-17.
Attribution. Source: LMArena (lmarena.ai), CC BY 4.0
Scores rounded; model entries mapped to LLM Reference models; snapshot 2026-09-22.
How to read this benchmark
This benchmark scores models where higher is better. Use scores for directional filtering and shortlisting, not universal quality ranking; then validate pricing, context window, provider availability, and fit for your workload.
Trust this score when
- The row names the exact model configuration, benchmark version, evaluator, harness, and observation date.
- Cost, token, and step averages are evaluated together with the score for the same configuration.
- You validate the configuration in your own Cursor workload before making a model choice.
Be cautious when
- Rows are collapsed into one score per base model or compared across benchmark versions.
- Small score differences are treated as decisive despite Cursor's variance caveat.
- A qualified row, including Grok 4.5's disclosed training-data advantage, is used as neutral winner evidence.
Related benchmarks
Last reviewed: Oct 17, 2026