The quality-per-credit frontier

How this report is built — read this before the tables

The quality scores are not ours. Every ranked model in the catalog is matched by name to an entry on Artificial Analysis, an independent benchmarking site that runs public human-preference arenas for image, video and speech models, and each row carries that arena's current ELO for the matched entry. The tables below show the exact Artificial Analysis leaderboard name each score was taken from, and each section links the specific arena. The credit prices are ours; the scores are theirs.

One honest caveat, stated here rather than in a footnote: the catalog stores each model's current arena score only — there is no score history. Every “frontier over time” view on this page is therefore constructed by grouping models by their release date and reading today's scores across those groups. It shows what each generation of models is worth now, not what any leaderboard looked like on a past date — older models' scores keep moving as new arena votes come in.

Price means one thing throughout: credits per job — the fewest credits one complete generation can cost on Versely, taken from each model's own price matrix. That is the only figure comparable across models that bill per second, per megapixel or flat per call. Scored models that publish only a per-1,000-character rate are listed with that rate labeled as such and kept out of the charts and the frontier. The pool is the published model catalog — one row per model page, so every row here is a model you can actually open and run.

What the data says

• Text-to-video: the 2025 cohort peaks at 1235 ELO, the 2026 cohort at 1336. Outscoring the 2025 peak now starts at 6 credits per job (Wan 3.0 Text to Video).
• Image-to-video: the 2025 cohort peaks at 1296 ELO, the 2026 cohort at 1336. Outscoring the 2025 peak now starts at 34 credits per job (Happy Horse 1.1 Image to Video).
• Text-to-image: the 2024 cohort peaks at 890 ELO, the 2026 cohort at 1171. Outscoring the 2024 peak now starts at 1 credits per job (HiDream O1 Image).
• Image editing: the 2025 cohort peaks at 1104 ELO, the 2026 cohort at 1181. Outscoring the 2025 peak now starts at 2 credits per job (Grok Imagine Image 2.0).

Text-to-video

28 scored models · scores from the Artificial Analysis Text to Video arena

Every scored text-to-video model, by arena score

# is the model's position within Versely's published catalog for this arena, not its position on the full Artificial Analysis leaderboard.

Image-to-video

27 scored models · scores from the Artificial Analysis Image to Video arena

Every scored image-to-video model, by arena score

# is the model's position within Versely's published catalog for this arena, not its position on the full Artificial Analysis leaderboard.

Text-to-image

40 scored models · scores from the Artificial Analysis Text to Image arena

Every scored text-to-image model, by arena score

# is the model's position within Versely's published catalog for this arena, not its position on the full Artificial Analysis leaderboard.

Image editing

25 scored models · scores from the Artificial Analysis Image Editing arena

Every scored image editing model, by arena score

# is the model's position within Versely's published catalog for this arena, not its position on the full Artificial Analysis leaderboard.

Text-to-speech

9 scored models · scores from the Artificial Analysis Text to Speech arena

This arena's pool is too small to cut into release-year cohorts honestly — fewer than two cohorts reach 3 priced models — so it gets the full ranking only, no trend claims.

Every scored text-to-speech model, by arena score

# is the model's position within Versely's published catalog for this arena, not its position on the full Artificial Analysis leaderboard. 9 of the 9 rows publish no complete-job minimum — their credit figure is the headline rate, labeled, and they are excluded from the chart and the frontier.

Frequently asked questions