跪拜 Guibai
← All articles
Backend · Java · Developer

SenseTime’s SenseNova Opens a Free, Multimodal API During Public Beta

By SimonKing ·
Read original on juejin.cn ↗ Google Translate ↗ Alt translation

A major Chinese AI lab is offering zero-cost API access to multimodal models with a straightforward rate limit and no billing friction. For developers testing agent integrations or multimodal pipelines, this removes the usual paywall that gates experimentation.

Summary

SenseNova is a Model-as-a-Service platform built on SenseTime’s own SenseCore infrastructure, offering general, multimodal, and conversational models through a standard API. The 2026 Token Plan removes all cost barriers during the beta: registration alone gets you an API key and a 1,500-call allowance refreshed every five hours, with no credits that expire or hidden charges.

The model lineup includes native multimodal models that handle documents, images, and tables at the architecture level, plus text-to-image models. A walkthrough shows how to wire the API into OpenCode — an agent coding tool — by setting the base URL and key, then optionally installing custom tools to unlock image generation directly inside the agent.

Output quality from the text-to-image models is solid enough for prototyping, and the free tier’s rate limit is generous for individual developers experimenting with agent-driven multimodal workflows.

Takeaways
SenseNova’s Token Plan public beta provides 1,500 free API calls every 5 hours across all models, with no prepayment or expiring credits.
The platform exposes general, multimodal, and conversational models — multimodal support is architectural, not a thin wrapper on a text model.
Registration at platform.sensenova.cn/docs yields an API key immediately; the base URL for calls is https://token.sensenova.cn/v1.
OpenCode can be configured manually by adding SenseNova as a provider with the API key and base URL, or by feeding the docs to OpenCode for auto-configuration.
Text-to-image models like sensenova-u1.5-lite and sensenova-u1-fast require a custom tool install inside OpenCode before they can generate images.
Image generation quality in testing was described as quite good for a free beta service.
Conclusions

The 5-hour refresh window on the free tier is unusual — most free API plans use daily or monthly caps. A shorter cycle suits rapid prototyping sessions where a developer burns through calls, waits a few hours, and resumes.

Native multimodal architecture means the model reads documents, images, and tables without pre-processing pipelines, which reduces integration complexity compared to stitching separate vision and text models together.

SenseTime is coupling free access with an agent ecosystem play: the walkthrough targets OpenCode, not a raw REST client, signaling that the free tier is meant to pull developers into agent-based tooling rather than one-off API experiments.

Concepts & terms
SenseNova Token Plan
SenseTime’s public beta program that makes all SenseNova model API calls free, with a quota of 1,500 requests refreshed every 5 hours and no payment required.
Native multimodal model
A model designed from the architecture level to process multiple input modalities — text, images, documents, tables — rather than bolting a vision encoder onto a text-only model.
OpenCode
An agent coding tool that can be configured with third-party model providers; it supports custom tool installation to extend capabilities like text-to-image generation.
Source: juejin.cn ↗ Google Translate ↗ Backup ↗