SenseNova-V on live video streams
SenseNova-V · SenseTime · VLM · API
SenseNova-V is SenseTime's proprietary vision-language flagship, offered as a hosted API rather than open weights, with broad perception coverage and particular strength in video comprehension. Parameter count is undisclosed. SenseNova-V is not currently in Overshoot's live model catalog; when a proprietary model like this is reachable through the API, it is as a passthrough, where Overshoot manages the stream, frame selection, and streaming transport while inference runs on the provider's infrastructure.
SenseNova-V is not currently in the live Overshoot model catalog (verified against the live model catalog on 2026-07-14). Availability changes over time - query GET /v1beta/models for the current list.
- Developer
- SenseTime
- Parameters
- Undisclosed
- Context window
- 128K tokens
- License
- Proprietary
- Released
- Apr 2025
- Inputs
- Text, images, video frames
- Overshoot availability
- Not in live catalogas of 2026-07-14
What SenseNova-V is good at
SenseNova-V is built for broad visual perception, covering general scene understanding, object and scene recognition, and detailed visual question answering across a wide range of everyday and specialized imagery. SenseTime has positioned it particularly around video comprehension, tracking context across a clip rather than treating each frame in isolation.
As a closed, proprietary model, SenseNova-V is not something you can download or self-host, but it is reachable through SenseTime's standard API. Overshoot exposes proprietary models like this as passthroughs, using the same chat-completions shape as every other model on the platform, though SenseNova-V itself is not currently in the live catalog.
- Broad general-purpose visual perception
- Strong multi-frame video comprehension
- Standard hosted-API access with no self-hosting required
SenseNova-V and the Overshoot workflow
SenseNova-V is not currently in Overshoot's live model catalog; GET /v1beta/models returns the current list. If a passthrough for it were added, the request shape would be identical to any other model on the platform: publish live video over WebRTC to open a Stream, then reference it with an ovs:// URL, the latest frame, an exact timestamp, or a recent segment sampled at 1 fps, inside a standard chat-completions request.
For any passthrough model, Overshoot manages the stream lifecycle, frame selection, and SSE streaming transport for the response, while the actual inference happens on the upstream provider's infrastructure, so overall latency depends on their serving conditions rather than Overshoot's hosted-model baseline.
SenseNova-V among proprietary vision models
SenseNova-V competes with other closed, API-only vision flagships such as Gemini 3 Pro, GPT-5, and Claude Sonnet 4.5. Compared with those, it is positioned around SenseTime's particular strength in video-native comprehension, and choosing between them usually comes down to which provider's perception behavior and pricing fit a given application best.
Frequently asked questions
Can SenseNova-V analyze live video?
SenseNova-V is built for multi-frame video comprehension, but it is not currently in Overshoot's live model catalog. Models that are live answer questions about a WebRTC stream through ovs:// frame references in a standard chat-completions request. The catalog changes over time, so check GET /v1beta/models.
Is SenseNova-V open source?
No. SenseNova-V is a proprietary, closed model with an undisclosed parameter count, reachable only as a hosted API from SenseTime rather than a self-hostable download.
How does SenseNova-V compare to Overshoot-hosted models?
Overshoot-hosted open-weight models typically answer in about 200ms because they run on Overshoot's own fast path. A proprietary model like SenseNova-V would be exposed as a passthrough, with latency depending on the provider's infrastructure. SenseNova-V is not currently in the live catalog; check GET /v1beta/models for what is.
What is SenseNova-V best used for?
SenseNova-V suits general-purpose visual question answering and, in particular, tasks that benefit from strong multi-frame video comprehension, such as summarizing or reasoning about what happened across a clip rather than a single image.