Qwen/Qwen3.8-Max-0902 · api.gmi-serving.com
Find any moment in a 90‑minute recording.
Ask in plain words. The search model reads timecoded contact sheets of the whole recording, zooms into the best windows as video and pins the exact seconds. Pick Qwen3.8-Max (looks; Gemini 3.8 Flash listens for it) or Gemini 3.8 Flash on its own. Every call runs on GMI Cloud with tokens, thinking, latency, speed and cost live on screen.
262Ktokens of context on Qwen3.8-Max
image + videoread by the search model; audio by the listener
160 of 634frames read to search a 10-minute film
$1.65 / $4.95per million tokens in / out on GMI (Max 0902 snapshot)
Drop a recording
Drop or click to browse. mp4, mov, mkv, webm or audio, up to 400 MB (sent in 4 MB pieces). Or paste a YouTube link below.
no file
The agent
idle
Drop a recording to start.
contact sheets (coarse pass)
candidate windows
fine pass evidence
the moment
- Plan the searchRead the duration, choose the sampling grain
- Sample the whole recordingFrames tiled into timecoded contact sheets
- Coarse passThe search model looks at every sheet; the audio model listens to every window; hits merge into candidates
- Fine passVideo, or frames plus audio, or frames alone, whichever the model accepts; exact seconds pinned
- Cut the clipFrame-accurate re-encode of just the moment
- Transcribe and translateThe audio model hears the clip: source cues plus three translations, names flagged
- DeliverSoft tracks muxed, one burned in, SRT and VTT
Build your own omni agent on GMI Cloud
One endpoint, 200+ models, OpenAI-compatible. Create a key and this whole pipeline runs on your account.
Build this on GMI Cloud
One endpoint. 200+ models. Pay per token, scale to zero.