{"id":"vemana","slug":"vemana","name":"Vemana","orgId":"udgam-labs","family":"vemana","releaseDate":null,"availability":"announced","openWeights":true,"license":null,"paramsTotal":null,"paramsSource":null,"contextWindow":null,"knowledgeCutoff":null,"modelCardUrl":"https://udgamlabs.com/blog/vemana","hfRepo":null,"modalities":["text"],"aliases":["Vemana","vemana","vemana-base","Vemana-Base","udgamlabs/vemana-base","vemana-base-gguf","udgamlabs/vemana-base-gguf","vemana 278m"],"aliasNote":null,"origin":"tracked","tracked":{"trackedSince":"2026-10-27","announced":"2026-10-27","blurb":"Udgam Labs' 278M-parameter Telugu base model, trained from scratch with its own Telugu tokenizer and released under Apache-2.0. It continues Telugu text (it does not follow instructions) and runs on a laptop CPU or in a browser tab.","api":"none","openWeights":"yes","params":{"text":"278M (278,214,144)","source":"Udgam Labs technical report; Hugging Face model card"},"facts":[{"label":"Type","value":"Base model: continues Telugu text; no instruction or safety tuning"},{"label":"Architecture","value":"Gemma 4 text layout (Gemma4ForCausalLM): 18 layers, width 1,024, 8 query and 2 key-value heads, full attention, tied embeddings"},{"label":"Tokenizer","value":"Its own 32,000-token Telugu tokenizer with byte fallback"},{"label":"Context","value":"1,024 tokens (the length it was trained at)"},{"label":"Training data","value":"5.1B unique tokens of public Telugu web text from Sangraha (AI4Bharat), FineWeb-2 and HPLT 3.0, read about 2.35 times (12.0B token positions)"},{"label":"Training run","value":"From random initialization on one H100 in about 16 hours, Oct 26-27, 2026, streamed live"},{"label":"Release","value":"Vemana-Base weights, every 10% checkpoint (Vemana-Checkpoints), GGUF Q8_0 (297 MB) and F16 (558 MB), and the evaluation code on GitHub"},{"label":"Browser demo","value":"Runs fully in the browser tab on udgamlabs.com (297 MB download, text stays on the device); also a Hugging Face Space"},{"label":"License","value":"Apache-2.0"}],"claims":["Lab claim: about $69 of GPU time for the training run.","Lab claim: 0.361 bits per byte on 416 Telugu Wikipedia pages from Sep-Oct 2026, ahead of every other open base model under 1B it tested, including nanochat-telugu-560M (0.383); Sarvam-1 (2.5B) scores 0.356.","Lab claim: 0-shot Belebele-te 25.1%, MILU-te 25.3%, IndicXNLI-te 35.0%, IndicCOPA-te 58.4%, which the lab reports as at or near chance.","Lab claim: about 65 tokens per second on a Mac mini CPU."],"sources":[{"kind":"lab","title":"Vemana, a small Telugu model trained live in public (Udgam Labs blog)","url":"https://udgamlabs.com/blog/vemana","date":"2026-10-27"},{"kind":"lab","title":"Vemana model page with in-browser demo (Udgam Labs)","url":"https://udgamlabs.com/models/vemana","date":"2026-10-27"},{"kind":"paper","title":"Vemana: a 278M Telugu language model trained from scratch (technical report)","url":"https://udgamlabs.com/research/vemana","date":"2026-10-27"},{"kind":"repo","title":"udgamlabs/Vemana-Base on Hugging Face","url":"https://huggingface.co/udgamlabs/Vemana-Base","date":"2026-10-27"},{"kind":"repo","title":"udgamlabs/vemana on GitHub","url":"https://github.com/udgamlabs/vemana","date":"2026-10-27"},{"kind":"lab","title":"Jitesh K Prasad (Udgam Labs) on X: Vemana release","url":"https://x.com/jiteshkprasad/status/2115025372723151272","date":"2026-10-27"}],"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 27, 2026; no independent results yet.","lead":"As of Nov 3, 2026, Vemana is pending benchmarks on isitsota: announced Oct 27, 2026, with no independent results on the boards we track yet.","confidence":"low"},"status":"PENDING_BENCHMARKS","verdicts":[]}