Cola2 vs Kimi K3

Cola2 20

Chat template 20, A chat template in tokenizer_config.json (4,116 characters), which came with the Qwen3 tokenizer. Tool-calling format 0, none. Trained tool calling 0, not trained for tool calling. Structured output 0, Nothing about JSON or schema-constrained output. Reasoning control 0, no reasoning mode. Context 0, no context length is stated. 20 + 0 + 0 + 0 + 0 + 0 = 20.

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.

Cola2 25

Memory 15, the only build is the F32 safetensors set, 35.07 GB, and 35.07 GB × 1.2 = 42.1 GB, within 64 GB. Loaded as BF16, as the card's example does, it needs about half, but the line reads the file size. Runtimes 0, none of the six runs it in a released version. Builds 0, one set of weights and no quantised build. Terms 0, no licence for the weights. 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 nothing names an API that serves it. 15 + 0 + 0 + 0 + 10 + 0 = 25.

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.

Cola2 40

File safety 15, safetensors only, no .bin, .pt or .pth files (25). Provenance 15, the files sit in the ByteDance-Seed organisation, which the ByteDance-Seed GitHub repository for the model links to (10). Safety evaluation 0, no safety results. Agent guidance 5, General cautions in the GitHub model card: not instruction-tuned or safety-aligned, not for use as an assistant, for facts or advice, or for high-impact decisions about people. Licence risk 0, With no licence for the weights there is nothing to hold ByteDance to. Security programme 5, the GitHub repository's SECURITY.md sends reports to ByteDance's security centre and says how they're assessed (5). 15 + 15 + 0 + 5 + 0 + 5 = 40.

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.

Cola2 58

Model card 8, Intended use and limitations (the out-of-scope list, and the unconverged latent model) and the architecture are on the card and the GitHub model card it links. Training data 15, the GitHub model card names the source (OLMo 2 pretraining text, retokenised) and the mix by stage with token counts, 346.03B in all. Last change 20, All six commits are from 6 November 2026, four days before this check, the last one updating a checksum in the README. Responsiveness 0, no discussions on the repository, so nobody has asked anything yet and the line reads as none answered, as settled in calibration. Version policy 5, A first release. File health 10, Config, files and card agree: F32 weights, a float32 config and a card that says to load them as bfloat16, with SHA256SUMS to check the download. 8 + 15 + 20 + 0 + 5 + 10 = 58.

Kimi K3 41

The model card earns 11, covering intended use, architecture, parameters and context, but not limitations, languages or a knowledge cut-off. Training data earns 5 for a general description with no cut-off. The last change earns 15, an encoding_k3.py update on 20 August, 74 days before 2 November and within 90; the GitHub repository's last commit is 6 August. Responsiveness earns nothing, with no Moonshot reply in the 40 most recent discussions. Version policy earns 5 for a first release. File health earns 5, because a report that the tokeniser goes past the declared vocabulary waits without a reply.