Devstral Small 2 vs Nemotron 3 Nano 4B

Devstral Small 2 46

Model card 11, Intended use, the architecture, the parameters and the context are on the card. Training data 0, Nothing on the card about the training data. Last change 10, the last commit, on 15 July 2026, updated the files to match mistral-common, 118 days before this check and within 180. Responsiveness 10, Of the 23 most recent discussions that aren't pull requests (9 December 2025 to 6 November 2026), 11 have a reply from a mistralai member, which is some. Version policy 5, Devstral Small 1.1 stays in its own repository, which this card links as the predecessor, with no dated note on the older card that we read. File health 10, Config, template, tokenizer and card agree, and the June and July commits fixed stray tokens and brought the files in line with mistral-common. 11 + 0 + 10 + 10 + 5 + 10 = 46.

Nemotron 3 Nano 4B 55

The model card earns 20, all seven items with limitations in the linked subcard. Training data earns 20, datasets listed with sources and dates, crawled sets with token counts, a cut-off and published NVIDIA datasets, though most synthetic sizes are undisclosed. Last change earns 5, as the files last changed on 20 March 2026, and TensorRT-LLM's Nemotron-H commits since 14 October touch MoE, NVFP4 and Puzzle paths this dense model doesn't use, in release candidates or no release. Responsiveness earns 0, with no NVIDIA reply to any of six questions, re-counted on 4 November. Version policy earns 5 as a first release. File health earns 5, since the LICENSE file is empty, reports and fixes wait without a reply, and the card's Transformers snippet names a repository that isn't public.

Devstral Small 2 60

File safety 25, safetensors only, no .bin, .pt or .pth files and no Python in the repository. Provenance 15, files in Mistral's verified mistralai organisation, which Mistral's site links to per our Mistral Large 4 listing (10). Safety evaluation 0, no safety results on the card. Agent guidance 0, the card has no limitations or safety section, and nothing on prompt injection, instruction priority or what the model refuses. Licence risk 15, Apache 2.0 has no use policy, threshold or change clause. Security programme 5, mistral.ai's security.txt gives a security contact and a submission policy, as read for the Mistral Large 4 listing (5). 25 + 15 + 0 + 0 + 15 + 5 = 60.

Nemotron 3 Nano 4B 65

File safety earns 15, safetensors and GGUF only but with 10 off because the documented commands pass trust_remote_code and load a parser plugin from the repository. Provenance earns 15, files in NVIDIA's linked organisation with a hash for every file. Safety evaluation earns 0, no results. Agent guidance earns 10, subcards that name prompt injection and the safeguards to add, without an instruction hierarchy. Licence risk keeps all 15, since the switch to the Nemotron licence happened on release day. The security programme earns 10 for the reporting route on the card and NVIDIA's public bulletins.

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.

Nemotron 3 Nano 4B 70

The chat template earns 20, a Jinja file in the repository. The tool-calling format earns 10 of 25, as only the template shows the format and the parser name appears only in a serving command. Trained tool calling earns 20, with tool-use benchmarks and data and several calls per turn in the template. Structured output earns 0, not claimed or documented. Reasoning control earns 10, thinking off per request with a parser that returns the reasoning apart. Context earns 10 for 262,144 tokens.