Agnes Video 2.5
Release date unpublished
Sapiens AI's video generation model, served through the Agnes AI platform's OpenAI Videos-compatible API. It supports text-to-video, first/last-frame keyframe control and multimodal reference generation.
- Price per second
- $0.025
- Max clip length
- 12 s
- Max resolution
- 2K
- Native audio
- —
- AA arena Elo
- 1066
- LMArena Elo
- —
- Capabilities
- text to video, image to video
- Licence
- Proprietary
Strengths & weaknesses
Strengths
- Cheap at $2.40/min at 1080p (Artificial Analysis listing; $0.040/s on the developer's docs)
- Text, keyframe and reference modes, with image, video and audio references
- Outputs from 720P to 2K across six aspect ratios
Weaknesses
- Clips run 4 to 12 seconds
- No published release date or changelog
Release day not published by the developer. Artificial Analysis lists 'Aug 2026', so releaseDate is null. The v2.0 docs say agnes-video-v2.0 was retired Sep 25, 2026 in favour of the 2.5 series (2.5 and 2.5 Flash). nativeAudio left null: the docs document audio only as a reference input ('Use audio or a video soundtrack as a reference'). A third-party ArcReel PR describes 2.5 audio as always-on, which is unconfirmed. The docs do not say which resolution is the default. AA's $2.40/min at 1080p matches the developer's $0.040/s 1080P rate. A sibling 'agnes-video-2.5-flash' (720P only) exists and is not an entry here. 'image-to-video' is mapped from the keyframe (first/last frame) and reference-image modes. Not on LMArena's video board. Notable false as of 2026-10-27: outside the AA-Video-T2V v2.1 top 10 (Elo 1066; #10 is 1128) and not on LMArena's video board.
Links
API model strings
- agnes-ai
agnes-video-2.5