SenseNova 日日新
SenseTime's multimodal agent models, open image models and a free-beta Token Plan
Overview
SenseNova (日日新) is SenseTime's foundation-model line: the SenseNova 6.8 Flash Lite agent model and U1.5 image models through a self-serve Token Plan API, per-token V6.5 models on SenseCore, and open-weight U1/U1.5 and SI models on GitHub and Hugging Face.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 14 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- 60,000 credits per 5 hours (special models excluded)
- SenseNova 6.8 Flash Lite and SenseNova U1 Fast
- Native Cowork-Skills support
- Quick connect with Hermes Agent and OpenClaw
- Up to 20 API keys
- No price or allowance published yet
- SenseCore postpaid, billed hourly
- SenseCore postpaid, billed hourly
- SenseCore postpaid, billed hourly
What you can produce with SenseNova 日日新
- SenseNova 6.8 Flash Lite multimodal agent model through an OpenAI-compatible API
- A Token Plan subscription, free in public beta (Lite and Pro tiers not yet priced)
- SenseNova U1.5 Lite and Fast image generation and editing models
- Open-weight SenseNova-U1/U1.5 (Apache-2.0) and SenseNova-SI spatial-reasoning models
- Per-token SenseNova V6.5 Pro and Turbo APIs on SenseCore
- DeepSeek, GLM-5.2 and Kimi K3 served on the same Token Plan
- The Raccoon (小浣熊) office agent app
SenseNova is no longer an enterprise-only buy. The sensenova.cn Token Plan is self-serve and free while in public beta, with 60,000 credits per rolling five hours, and it serves SenseTime's 6.8 Flash Lite agent model alongside DeepSeek, GLM-5.2 and Kimi K3. The catch is that paid Lite and Pro tiers are announced but unpriced, so budgeting means using SenseCore's per-token rates instead (V6.5-Pro at ¥3/¥9 per million).
The open-weight work is the stronger story: SenseNova-U1.5 is Apache-2.0 and SenseNova-SI is a credible spatial-reasoning release, though the "beats GPT-5" framing is narrower in its own paper, and the SuperCLUE 75.35 is vendor-relayed. Good for Chinese-market agent and image workloads; less useful for anyone who needs a stable price list today.
What we found
SenseNova (日日新) is the model line SenseTime unveiled at its Shanghai Tech Day in April 2023, and in 2026 it is sold three ways: a developer API platform at sensenova.cn, per-token billing on the SenseCore cloud, and open-weight releases on GitHub and Hugging Face. The current platform lineup is agent-oriented. SenseNova 6.8 Flash Lite, which reached the Token Plan as a preview on 11 August 2026, is a lightweight multimodal agent model aimed at data analysis, research reports and slide decks; SenseNova U1.5 Lite and Fast handle image generation and editing. The same Token Plan also serves third-party open models, including DeepSeek V4 Flash, GLM-5.2 and Kimi K3.
The Token Plan is free during its public beta. Each credit pool gets 60,000 credits per rolling five hours and 600,000 per rolling week, and SenseTime says paid Lite and Pro tiers are coming but has not priced them. Buyers who need a published rate card can use SenseCore instead, which lists SenseNova-V6.5-Pro at ¥3 per million input tokens and ¥9 per million output, and V6.5-Turbo at half that. Earlier versions of this page quoted a ¥0.58 rate that could not be traced.
The open-weight side is substantive. SenseNova-U1.5-8B-MoT, released on 20 August 2026 under Apache-2.0, generates and edits images without a separate VAE, using what SenseTime calls the NEO-unify architecture. SenseNova-SI is a spatial-reasoning model; its arXiv paper claims it beats GPT-5 on specific spatial abilities such as perspective-taking, which is narrower than the press release's headline. On benchmarks, SenseTime reports that SuperCLUE's December 2025 multimodal report ranked V6.5 Pro first among Chinese models at 75.35, a result relayed by the vendor.
Commercially, SenseTime reported RMB2.33 billion of generative-AI revenue in the first half of 2026, almost 80% of group revenue. Its Raccoon office agent names Lenovo, Ping An Technology and JD.com as enterprise users. Most outside coverage relays SenseTime's own announcements, so treat the benchmark and adoption claims as self-reported.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
About this listing
This entry was compiled from publicly available data including SenseNova 日日新's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with SenseNova 日日新 unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to SenseNova 日日新 directly →
Spotted something out of date? Suggest an update →
Alternatives to SenseNova 日日新
More in LLMs & Chat
