Devstral Small 2 vs LongCat-2.5

Devstral Small 2 90

Memory 25, the smallest build at 4 bits from a runtime's own library is Ollama's devstral-small-2 at Q4_K_M, 15 GB, and 15 GB × 1.2 = 18 GB, within 24 GB. Mistral's own FP8 files are 25.79 GB. Runtimes 25, Released vLLM 0.33.0, SGLang 0.5.23 and Transformers 5.22.2 have the mistral3 model, llama.cpp b12242 has the architecture with a GGUF from its own organisation, and Ollama's library has devstral-small-2. Builds 10, one quantised precision from Mistral, the FP8 weights it releases. Terms 15, Apache 2.0 in the card's metadata and text, an OSI licence with 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 5, Mistral's own API serves Devstral Small 2: the card says the model is available through its API and points there when a local runtime underperforms. 25 + 25 + 10 + 15 + 10 + 5 = 90.

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.

Devstral Small 2 65

Chat template 20, chat_template.jinja in the repository (11,362 characters), with the tokenizer's tekken file for mistral-common. Tool-calling format 25, the card documents tools in the OpenAI function format with an example, names the mistral tool-call parser with auto tool choice for vLLM, shows the same for SGLang, and ships test prompts with the expected tool-call output. Trained tool calling 10, the card says the model excels at using tools, and its tests cover one tool call and calls made one at a time in sequence. Structured output 0, Nothing on the card about JSON or schema-constrained answers for this model. Reasoning control 0, no reasoning mode is documented, and a June 2026 commit removed stray thinking tokens from the files. Context 10, 256K tokens per the card, served at 262,144 in its vLLM command. 20 + 25 + 10 + 0 + 0 + 10 = 65.

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.