Cosmos-Reason2-2B vs MiniCPM-V 4.7 35B-A3B

Cosmos-Reason2-2B 39

The card misses languages and a cut-off (14). Training data is named by dataset without sizes or a cut-off (5). The Hub repository last changed on 30 April 2026, 188 days before this check, which alone gives 5. NVIDIA's GitHub repository last changed with a README redirect to Cosmos 3 on 7 June 2026, 150 days back, and this run counts the publisher's documentation commits on this line (10). None of its 3 discussions has an NVIDIA reply, counted by author flags on 4 November (0). The 2B's weights were replaced in place in March with a dated line, and the family is now superseded by Cosmos 3 with a dated note (5). The gated files couldn't be checked against the card, which itself mixes float16 and BF16 and links a newer licence than the gate shows (5). 14 + 5 + 10 + 0 + 5 + 5 = 39.

MiniCPM-V 4.7 35B-A3B 68

Model card 8, it covers intended use, architecture and parameters but not limitations, context length, languages or a knowledge cut-off. Training data 0, none described. Last change 20, a commit on 3 November 2026. Responsiveness 20, OpenBMB answered 8 of 12 discussions. Version policy 10, earlier versions keep their repositories and the changes are dated and described. File health 10, the files agree. 8 + 0 + 20 + 20 + 10 + 10 = 68.

Cosmos-Reason2-2B 65

Memory earns the full 30 because the BF16 file is 4.9 GB, about 5.9 GB with the fifth added, though NVIDIA itself asks for 24 GB of GPU memory. Runtimes get 20 because the model keeps Qwen3-VL's architecture, which Transformers, vLLM, SGLang and llama.cpp run in released versions, while LM Studio's catalogue has only the base. Builds score 0, as NVIDIA publishes no quantised build of the 2B. The NVIDIA Open Model License allows free commercial use under a use policy (10). The gate is an automatic click-through that needs an account (5), and nothing hosts the 2B (0).

MiniCPM-V 4.7 35B-A3B 90

Memory 25, since OpenBMB's IQ4_XS GGUF with its vision projector is 19.6 GB and needs about 23.6 GB with the fifth added, which fits 24 GB, while its Q4_K_M GGUF, 4-bit safetensors builds and Ollama entry need 32 GB. Runtimes 25, from vLLM 0.33.0, SGLang 0.5.23, Transformers 5.18.0 and later, llama.cpp's own MiniCPM-V 4.7 support and OpenBMB's entry on Ollama. Builds 15, OpenBMB publishes FP8, three 4-bit safetensors builds and GGUF. Terms 15, Apache 2.0 with nothing added. Download 10, nothing is gated. Hosted 0, no inference provider is mapped. 25 + 25 + 15 + 15 + 10 + 0 = 90.

Cosmos-Reason2-2B 60

File safety gets the full 25, one safetensors file loaded by Transformers' own code with a template, no remote code. Provenance gets 15 for files in NVIDIA's own organisation with hashes. The licence has a use policy and guardrail clause and NVIDIA can revoke or update it, so licence risk keeps 5 of 15. NVIDIA's PSIRT route and bulletins give the security programme 10. Agent guidance gets 5 for general cautions about guardrails and physical risk, and safety evaluation 0, with no safety results.

MiniCPM-V 4.7 35B-A3B 55

File safety 25, safetensors and GGUF only, no Python in the main repository and no remote code needed. Provenance 15, the files come from the openbmb organisation OpenBMB's GitHub README links to, with a SHA-256 for every weight file. Safety evaluation 0 and agent guidance 0, nothing published on either, and the card has no limitations section. Licence risk 15, plain Apache 2.0. Security programme 0, no security.txt or SECURITY.md.

Cosmos-Reason2-2B 65

A Jinja template ships as chat_template.json (20). Tools are declared only by the template it inherits from Qwen3-VL, with no NVIDIA documentation or parser (10), and nothing shows the 2B was trained for tool use (0). NVIDIA's prompt files ask for JSON with named keys for boxes, events and trajectories (15). Reasoning is asked for in the prompt and vLLM's qwen3 parser returns it apart (10). The card states 256K input tokens (10).

MiniCPM-V 4.7 35B-A3B 70

Chat template 20, a Jinja file in the repository. Tool-calling format 10, the card shows the tools array going in but not the call coming back, and names no parser. Trained tool calling 20, the template renders several calls in one turn. Structured output 0, not documented or claimed. Reasoning control 10, enable_thinking per request, off by default, with the trace between markers. Context 10, 262,144 tokens native.