{"id":"udgam-384m","slug":"udgam-384m","name":"Udgam-384M","orgId":"udgam-labs","family":"udgam-384m","releaseDate":null,"availability":"announced","openWeights":true,"license":null,"paramsTotal":null,"paramsSource":null,"contextWindow":null,"knowledgeCutoff":null,"modelCardUrl":"https://udgamlabs.com/blog/udgam-384m","hfRepo":null,"modalities":["text"],"aliases":["Udgam-384M","udgam-384m","Udgam 384M","Udgam-384M-Chat","udgamlabs/Udgam-384M-Chat"],"aliasNote":null,"origin":"tracked","tracked":{"trackedSince":"2026-10-24","announced":"2026-10-10","blurb":"Udgam Labs' 384M-parameter English model for tool calling inside apps, on the device itself: it turns requests into function calls, documents into JSON and tasks into shell commands for a person to review. Trained from scratch on a Gemma 4-compatible architecture and released under Apache-2.0.","api":"none","openWeights":"yes","params":{"text":"384M (384.4M; 20 layers, hidden size 1,024)","source":"Udgam Labs model page and Hugging Face card"},"facts":[{"label":"Architecture","value":"Decoder-only transformer compatible with Gemma 4 (Gemma4ForCausalLM), sliding-window plus global attention"},{"label":"Training","value":"Trained from scratch (random initialization) on roughly 8B tokens of licensed public and synthetic text, then supervised fine-tuning for chat and tool use"},{"label":"Context","value":"4,096 tokens (2,048 for the base model)"},{"label":"Language","value":"English"},{"label":"Formats","value":"safetensors, GGUF (Q8_0 410 MB; Q4_K_M 238 MB for phones) and MLX (4-bit build 216 MB); runs offline in Ollama, LM Studio, llama.cpp, MLX and transformers"},{"label":"Release","value":"Udgam-384M-Chat plus Udgam-384M-Base and Udgam-384M-Base-4K on Hugging Face; training code and data not released"},{"label":"License","value":"Apache-2.0 (weights and tokenizer)"}],"claims":["Lab claim: on Udgam's own 274-item held-out sets (which follow this model's prompt conventions), 75% right call/no-call decisions on relevance, 56% schema-valid JSON for unseen schemas and 85.9% right shell behaviour.","Lab claim: BFCL v3 simple 60, IFEval prompt-level strict 26.1; Udgam reports Granite 4.0 350M (90.7 / 47) and Qwen3-0.6B (80 / 58) ahead on both in the same harness.","Lab claim: XSTest declines 54% of clearly harmful prompts and over-refuses 15% of safe ones; released as not safety-hardened.","Lab claim: 444 tokens/s generation at Q4_K_M on an Apple M6 Mac mini GPU."],"sources":[{"kind":"lab","title":"Udgam-384M, a small model for tool calling inside apps (Udgam Labs blog)","url":"https://udgamlabs.com/blog/udgam-384m","date":"2026-10-10"},{"kind":"lab","title":"Udgam-384M model page (Udgam Labs)","url":"https://udgamlabs.com/models/udgam-384m","date":"2026-10-10"},{"kind":"repo","title":"udgamlabs/Udgam-384M-Chat model card (Hugging Face)","url":"https://huggingface.co/udgamlabs/Udgam-384M-Chat"},{"kind":"lab","title":"Udgam Labs on X","url":"https://x.com/udgamlabs/status/2108829455204683993","date":"2026-10-10"}],"firstBenchmark":null},"organization":{"id":"udgam-labs","name":"Udgam Labs","slug":"udgam-labs","country":null,"url":"https://udgamlabs.com","color":"#11110F","mark":"U"},"price":null,"answer":{"word":"Pending benchmarks.","scope":"Announced Oct 10, 2026; no independent results yet.","lead":"As of Nov 3, 2026, Udgam-384M is pending benchmarks on isitsota: announced Oct 10, 2026, with no independent results on the boards we track yet.","confidence":"low"},"status":"PENDING_BENCHMARKS","verdicts":[]}