LongCat-2.5 vs Qwen3.6-35B-A3B

LongCat-2.5 35

Memory 0, the smallest build is Meituan's W4A4 repository, 1,336 GB, and 1,336 GB × 1.2 = 1,603 GB, far above 256 GB. The BF16 files are 3,562 GB. Runtimes 0, none of the six runs it in a released version: vLLM 0.33.0, SGLang 0.5.23, Transformers 5.22.2 and llama.cpp b12242 have no LongcatCausalLM or pro_v code, and the model code config.json names isn't in the repository. Builds 10, one quantised build from Meituan, the W4A4 repository (4-bit experts, 1,336 GB). Terms 15, MIT in the metadata, the LICENSE file and the card, with nothing added. Download 10, the Hub API reports gated false for both repositories 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. 0 + 0 + 10 + 15 + 10 + 0 = 35.

Qwen3.6-35B-A3B 85

Memory earns 20 because the smallest 4-bit build from a runtime's own library is Ollama's Q4_K_M at 23 GB, which comes to 27.6 GB with the fifth added, so it needs a 32 GB machine rather than a 24 GB one. Runtimes earn the full 25, since released llama.cpp, Ollama, vLLM, SGLang and Transformers all run the architecture. Builds earn 10 for Alibaba's one quantised build, FP8, with no 4-bit or GGUF from the publisher. Terms earn 15 for plain Apache 2.0, download earns 10 for ungated files, and hosted earns 5 because DeepInfra, Featherless AI and Scaleway serve it through Hugging Face.

LongCat-2.5 46

Model card 11, Intended use, limitations (the usage considerations), the architecture and the parameter counts are on the card. Training data 5, A general description of the training stages and where the reinforcement learning tasks came from, with no sources, sizes or cut-off. Last change 20, the last commit, on 10 November 2026, updated the card, on the day of this check. Responsiveness 0, Neither of the 2 discussions that aren't pull requests (9 and 10 November 2026) has a reply from a meituan-longcat member, counted on 10 November by author flags and dates with a script, text not read. Version policy 5, LongCat-2.0 stays in its own repositories and 2.5 came out in new ones, with no dated note on the older ones that we read. File health 5, the template is in both places and agrees, but config.json's auto_map names configuration_longcat_oe and modeling_longcat_oe, which aren't in the repository, and the tokenizer config's 131,072-token length is a quarter of the config's 524,288 positions. 11 + 5 + 20 + 0 + 5 + 5 = 46.

Qwen3.6-35B-A3B 31

The model card earns 11, covering intended use, architecture, parameters and context but not limitations, languages or a cut-off. Training data earns 5 for a general description with no cut-off. The last change earns 5, a README edit on 24 April 2026, 192 days before this check, past 180 days and within a year, and Qwen's GitHub repository holds only a README and licence, so no reference code of Alibaba's changed later. Responsiveness earns 0, with no Qwen reply in the 40 most recent of 62 discussions that aren't pull requests, recounted on 2 November. Version policy earns 5, the earlier version kept in its own repository without a note. File health earns 5, the files agree but reports about tool calling and the template wait unanswered.

LongCat-2.5 60

File safety 25, safetensors only, no .bin, .pt or .pth files and no Python in the repository. Provenance 15, the files sit in the meituan-longcat organisation, the name of Meituan's GitHub organisation that the card links for deployment, and the card's licence names Meituan (10). Safety evaluation 0, no safety results on the card. Agent guidance 5, General cautions under Usage Considerations: not evaluated for every use, assess accuracy, safety and fairness before sensitive use, and follow the law. Licence risk 15, MIT with no use policy, no threshold and no right to change the terms. Security programme 0, no way to report a flaw on the card beyond a general contact, and Meituan's own security pages weren't looked up, so this scores as absent and is an open question. 25 + 15 + 0 + 5 + 15 + 0 = 60.

Qwen3.6-35B-A3B 55

File safety earns 25 for safetensors only, loaded by the runtimes' own code and a Jinja template with nothing run from the repository. Provenance earns 15, files in the Qwen organisation the publisher links and a hash for every file. Safety evaluation and agent guidance earn 0, since the card has neither safety results nor any guidance on refusals, injection or safeguards. Licence risk keeps all 15, as Apache 2.0 carries none of the four conditions and the late LICENSE file only wrote down what the card already said. The security programme earns 0, with no reporting route found on qwen.ai, the card or the GitHub repository, which is noted as unchecked for Alibaba's group-wide centre.

LongCat-2.5 75

Chat template 20, chat_template.jinja in the repository (11,823 characters), with a copy in tokenizer_config.json. Tool-calling format 25, the card documents how tools are declared (the OpenAI function format) and how calls come back (tool_calls with an id, a name and JSON arguments, answered by a tool message), in a worked example against an OpenAI-compatible server. Trained tool calling 10, the card describes reinforcement learning on long agentic tasks and shows one tool call in a turn. Structured output 0, Nothing on the card about JSON or schema-constrained output. Reasoning control 10, enable_thinking on or off per request through chat_template_kwargs, with the reasoning returned as reasoning_content, both shown on the card. Context 10, the card states no context length, but config.json gives 524,288 positions and the card's own evaluations ran at 512K and 1M sequence lengths after what it calls long-context training, so 128K or more is taken as trained. 20 + 25 + 10 + 0 + 10 + 10 = 75.

Qwen3.6-35B-A3B 70

The chat template earns 20 as a Jinja file in the repository. The tool-calling format earns 10, because the card names the qwen3_coder parser but leaves the return format to the template. Trained tool calling earns 20, with several calls to a turn in the template. Structured output earns 0, as JSON output isn't documented for the model. Reasoning control earns 10, thinking off per request with enable_thinking and the reasoning returned in think tags for the qwen3 parser. Context earns 10 for 262,144 tokens native.