Kimi K3 vs SenseNova-Loop-9B-Preview

Kimi K3 65

File safety earns 15, safetensors only with no pickle files, less 10 because the documented loading path runs Moonshot's Python with trust_remote_code and builds prompts with encoding_k3.py. Provenance earns the full 15, files in the organisation Moonshot links to with a SHA-256 for every shard. Safety evaluation earns 20 for the cyber assessment the UK AI Security Institute and CAISI ran and published, which found its safeguards didn't stop exploit development. Agent guidance earns 5 for the blog's general cautions about proactiveness and passing back thinking. Licence risk earns 10, losing 5 for the model-as-a-service revenue threshold and nothing else. The security programme earns nothing, with no security.txt, disclosure policy, SECURITY.md or public advisories found.

SenseNova-Loop-9B-Preview 50

File safety 15, safetensors only, no .bin, .pt or .pth files (25). Provenance 15, the files sit in the sensenova organisation, which the GitHub repository for the model links to (10), and the Hub gives a SHA-256 for the large files (5). Safety evaluation 0, no safety results on the card. Agent guidance 0, the card has no limitations or safety section, and nothing on instructions inside tool results or images. Licence risk 15, Apache 2.0 with no use policy, no threshold and no right to change the terms. Security programme 5, SenseTime's about page gives a security contact (5). 15 + 15 + 0 + 0 + 15 + 5 = 50.

Kimi K3 50

Memory earns nothing, because the only build of 4-bit precision or better from Moonshot or a runtime's own library is Moonshot's MXFP4 release of 1,561 GB, about 1,873 GB with the fifth added, far above 256 GB. Ollama and LM Studio list it only as a cloud model. Runtimes earn 15, for model code in released vLLM, SGLang and llama.cpp; Transformers has no kimi_k3 code of its own and loads Moonshot's remote code, so it doesn't count. Builds earn 10 for one publisher precision, the MXFP4 files the model was trained for. Terms earn 10, a custom MIT-based licence that is free for commercial use below a revenue threshold. Download earns 10, with no gate and no account. Hosted earns 5, since Moonshot's API serves it and five inference providers are live for it on Hugging Face.

SenseNova-Loop-9B-Preview 40

Memory 15, the only build is the safetensors set, 38.14 GB with its MTP head, since the Hub counts 9.41B of its parameters in F32. Runtimes 0, none of the six runs it in a released version. Builds 0, SenseNova publishes one set of weights and no quantised build. Terms 15, Apache 2.0 in the card's metadata and on the card, which links the LICENSE in the GitHub repository, and nothing added. Download 10, the Hub API reports gated false on 10 November 2026, so the files download with no account and nothing to accept. Hosted 0, Hugging Face maps no inference provider to it, and the card names no API that serves it. 15 + 0 + 0 + 15 + 10 + 0 = 40.

Kimi K3 65

The chat template earns 10, because there is no Jinja template and the prompt format exists only as encoding_k3.py, code you run. The tool-calling format earns 10, since Moonshot documents the tools channel but the parser name comes only from vLLM's and SGLang's own recipes. Trained tool calling earns 20 for indexed parallel calls in one turn. Structured output earns 5, as the card names it among its linked guides without saying how to ask the weights for it, and the how sits in hosted-API docs. Reasoning control earns 10, with per-request effort documented on the card and reasoning returned as reasoning_content. Context earns 10 for 1,048,576 tokens trained in.

SenseNova-Loop-9B-Preview 60

Chat template 20, chat_template.jinja in the repository (7,756 characters). Tool-calling format 10, only the template says it: a tools block, and calls written as tool_call blocks with function and parameter tags. Trained tool calling 10, the card says the training recipes include tool trajectories and that the model is for agentic workflows. Structured output 0, Nothing on the card about JSON or schema-constrained output. Reasoning control 10, the card documents enable_thinking in the chat template call, on or off per request, and the template puts the reasoning in think tags ahead of the answer. Context 10, 262,144 tokens in training, per the card's specification table, and config.json gives the language model 262,144 positions. 20 + 10 + 10 + 0 + 10 + 10 = 60.