Upload a Video or Audio File and Ask AI About It: Steps, Limits and Real Cost
To ask AI about a video or recording, open the attachment menu, choose "Video / audio", pick the file and type your question. You don't need to choose a model. On AIWITH.CHAT, a 30-minute, 50 MB recording costs about 24,000 beans per question and a 10-minute, 100 MB video about 32,000. Every follow-up reads the file again, so it costs about the same.
A quick note: we publish AIWITH.CHAT, the multi-model chat service used as the example below. Other tools handle the upload in a similar way, but their limits and pricing differ.
What one question costs
| File | Per question | Questions on the $9.9 monthly plan (3.2M beans) |
|---|---|---|
| 1-minute, 10 MB video | about 5,300 beans | about 600 |
| 5-minute, 8 MB recording | about 5,800 beans | about 550 |
| 30-minute, 50 MB recording | about 24,000 beans | about 130 |
| 10-minute, 100 MB video | about 32,000 beans | about 98 |
How this is calculated: Gemini 3.6 Flash at the October 2026 rates of 0.38 beans per input token and 1.88 per output token. Video is about 100 tokens per second and audio about 26. One reply is estimated at 1,200 tokens, plus a file traffic fee of 74 beans per MB.
You don't have to work this out yourself. Once the upload finishes, the attachment card shows an estimate such as "~24K beans this message", and you are charged for what the model actually uses. If you have a more expensive model selected that can also read media, that model's rate applies and the estimate changes with it.
How to upload: three short steps
- In the message box, open the attachment menu, choose "Video / audio" and pick a file.
- Wait for the upload progress to finish. The card shows the length and the estimated cost.
- Type your question and send.
If your current model can't read video or audio (DeepSeek, GPT, Kimi and Claude can't here), that message is answered by Gemini 3.6 Flash automatically. The file plays inside the message, so you can drag the progress bar to check a moment the answer mentions.
Three ways to spend less
Ask everything in the first message. You don't need to upload again to ask "what happens at 0:05?", but the model reads the file again and you pay for it again. Three follow-ups on a 10-minute video add about 97,000 beans. Put it all in one message instead:
Summarize this recording, list each person's action items and deadlines, and tell me at what minute the longest disagreement starts.
Trim before you upload. If you only care about two minutes, upload those two minutes.
Send audio when the picture doesn't matter. Lectures, interviews and podcasts rarely need the video track. Ten minutes of audio costs about 9,400 beans on the same rates, against 32,000 for ten minutes of video.
What to upload
- Phone recording, meeting capture, local video: upload it as described here.
- Online videos such as YouTube or TikTok clips: save the file, then upload it.
Limits
| Limit | Video | Audio |
|---|---|---|
| Formats | mp4, mov, webm | mp3, m4a, aac, wav, ogg |
| Max size | 100 MB | 50 MB |
| Max length | 10 minutes | 30 minutes |
Each message takes one file; send it, then add the next one. Each account can upload up to 2 GB per day. Files are kept for 180 days, after which the message shows "File expired".
How accurate is it?
In our tests with public sample files, Gemini 3.6 Flash matched what happens in a 13-second video, and transcribed a 16-second recording sentence by sentence with time ranges. That is a small sample. Very short clips are where speech recognition slips most easily, so for names, amounts and dates, ask for the timestamp and listen to the original.
Try it on a recording you already have
Open the Gemini page on AIWITH.CHAT, upload the file and ask:
Summarize the key points of this recording with a timestamp for each point.
Upload a file on the Gemini page