SenseNova · Download verified

SenseNova Loop 9B Preview

Run a downloadable text-and-image model for document questions, coding and tool-assisted work. Its unusual design needs compatible serving software.

Released 2026-10-30 · First reviewed 2026-11-02 · Checked 2026-11-03.

We still need comparable independent tests. You can compare the confirmed features and limits below, but we cannot yet give this version an Agent Fit or Practical Value rating. Try it on your hardware and tasks before relying on it.

Runs onYour hardwareHardware and setup costs apply

Documented capacityNot verifiedSee input limits below

What you can use today

You download the model files and supply the hardware and software to run them. Capacity is the amount of information a request can contain; settings and available memory can limit what fits.

Exact download ID: sensenova/SenseNova-Loop-9B-Preview. The model reuses some of its layers twice while producing an answer. This is a fixed part of its design, separate from the optional thinking toggle. Publisher benchmark results do not establish independent quality or speed here.

Apache 2.0 project license and verified weight shards plus custom model code. The documented environment uses Transformers 5.18.0, PyTorch 2.11.0 and CUDA 13.0. SGLang and vLLM instructions require the specified implementation branches; a generic install command is not enough. Check memory for your actual input length.

The card describes training sequences up to 262,144 tokens, and the configuration uses that position limit. A universally supported serving limit and maximum answer length are not established. The 256-token example is not a model ceiling.

No verified hosted token tariff or local operating-cost measurement. Include hardware and software maintenance when comparing it with an online API.

Provider release and specifications ↗

Check availability and plans ↗

Provider documentation ↗

More setup and licensing details ↗

Test it before switching

Give each model the same examples, tools and definition of success. Record correct outcomes, response time, retries and total cost. Include the time and hardware needed to run it.

Compare caching and long-context costs ↗