Cosmos-Reason2-2B vs LFM3-1.7B

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).

LFM3-1.7B 100

Chat template 20, chat_template.jinja in the repository (12,129 characters), with roles, tools, deferred tools, tool results and the thinking switch. Tool-calling format 25, the card and Liquid's tool-use page document tools passed through tools= in apply_chat_template, calls returned as XML and results sent back as tool messages. Trained tool calling 20, the template loops over every entry in an assistant turn's tool_calls, so several calls fit in one turn, and the card says the model was trained on tool calls and agentic traces and with reinforcement learning inside agent harnesses. Structured output 15, Liquid's API guide documents response_format with a JSON Schema, with lfm3-1.7b as the model in its examples, and says the schema is checked before generation starts. Reasoning control 10, the card documents enable_thinking=False in the chat template to turn reasoning off per request, with reasoning on by default inside think tags, and preserve_thinking for earlier turns. Context 10, 256K tokens per the card, which says the mid-training stage extended the context window to 256K, and max_position_embeddings 262144 in config.json. 20 + 25 + 20 + 15 + 10 + 10 = 100.

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 The 2B's weights were replaced in place in March with a dated line, and the family is now superseded by Cosmos 3 with a dated note (5).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.

LFM3-1.7B 67

Model card 17, Intended use (local agents, tool calling, structured outputs, fine-tuning), a limitation (not recommended for agentic coding or complex multi-step reasoning), architecture (24 layers: 13 gated short convolutions, 5 sliding-window and 6 global attention), parameters (1.73B), context (256K) and 16 languages. Training data 5, A general description: 22T tokens of pre-training in two stages with the kinds of data added since LFM2, 1.75T tokens of supervised fine-tuning, then mixed reinforcement learning, distillation from six specialist teachers and agentic reinforcement learning. Last change 20, the last commit, on 9 November 2026, added the OpenVINO checkpoint to the card's inference table, a day before this check. Responsiveness 10, Of the 22 discussions that aren't pull requests (27 October to 8 November 2026), 7 have a reply from a LiquidAI member, which is some. Version policy 5, LFM3 came out in new repositories and the LFM2.5 models stay in theirs. File health 10, Config, template and card agree on the lfm3 architecture, the 262,144-token context and the thinking switch. 17 + 5 + 20 + 10 + 5 + 10 = 67.

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).

LFM3-1.7B 85

Memory 30, Liquid's Q4_K_M GGUF is 1.10 GB, and 1.10 GB × 1.2 = 1.3 GB, which fits 8 GB, well inside the 16 GB line. Runtimes 15, Released llama.cpp has the lfm3 architecture (src/models/lfm3.cpp at b12242), Ollama's library has an lfm3 entry, and LM Studio's catalogue runs the architecture (its liquid/lfm3-vl-2b entry; lmstudio-community publishes a GGUF of this model). Builds 15, Liquid's GGUF repository holds Q4_0, Q4_K_M, Q8_0, F16 and BF16 files, and Liquid publishes 4-bit and 8-bit MLX builds and an ONNX repository, all in the LiquidAI organisation. Terms 10, the LFM Open License v1.0, a custom licence that lets anyone use it commercially for free below a threshold of $10M annual revenue. Download 10, the Hub API reports gated false for the model and its GGUF, MLX and ONNX repositories on 10 November 2026, so the files download with no account and nothing to accept. Hosted 5, Liquid's own API lists lfm3-1.7b among its models, so the publisher hosts it. 30 + 15 + 15 + 10 + 10 + 5 = 85.

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.

LFM3-1.7B 40

File safety 15, safetensors only, no .bin, .pt or .pth files (25). Provenance 15, files in the LiquidAI organisation, which Liquid's docs page for the model and its blog post link to (10). Safety evaluation 0, no safety results on the card or in Liquid's blog post for the LFM3 family, neither of which mentions safety testing. Agent guidance 0, Nothing on prompt injection, an instruction hierarchy, what refusals cover or safeguards to add. Licence risk 10, 15 less 5 for the $10M annual revenue threshold on commercial use. Security programme 0, no way to report a flaw found. 15 + 15 + 0 + 0 + 10 + 0 = 40.