Together AI

3 models measured.

deepseek-ai/DeepSeek-V4.1-Flash-1017

Images: 839 “Multimodal payload is 1503781440 bytes (1434.1 MiB) which exceeds the gRPC single-message ceiling of 1502189977 bytes (1432.6 MiB). Reduce the number of images, lower their resolution (e.g. `detail=”
768×432 frames: ≥ 384
Tokens / frame: 223
Inline body: ≥ 32 MB
Images by URL: ≥ 128 MB
Video file: No video “Invalid multimodal request: invalid media request for model deepseek_v41: modality video is not supported by model spec deepseek_v41”
Video tok/min: No “Invalid multimodal request: invalid media request for model deepseek_v41: modality video is not supported by model spec deepseek_v41”
Frames/min: —
Video input: No
Tool messages: Strict order
Extra fields: All accepted

Together caps the decoded media in one request at 1,502,189,977 bytes (≈1.43 GB). Images are decoded to tensors before this check, so pixel count matters, not file size.

Measured 2026-10-31

zai-org/GLM-5.3-Flash

Images: 1,004 “Failed to start generation: At most 999 image(s) may be provided in one prompt. (parameter=image)”
768×432 frames: ≥ 384
Tokens / frame: 450
Inline body: ≥ 32 MB
Images by URL: ≥ 128 MB
Video file: 256 MB “Could not load the media in this request — video #1: video too large: 320000000 exceeds limit 268435456.”
Video tok/min: 952
Frames/min: —
Video input: Yes
Tool messages: Strict order
Extra fields: All accepted

Together caps the decoded media in one request at ≈1.43 GB (seen on DeepSeek V4.1 Flash; likely shared).

Measured 2026-10-31

moonshotai/Kimi-K3

Accepts a video_url with HTTP 200, but the video adds no tokens: the model never sees it.
Together caps the decoded media in one request at ≈1.43 GB (seen on DeepSeek V4.1 Flash; likely shared).

Measured 2026-10-31