CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins88% CONFIDENCEOverall confidence: 88%How well the pin's source and references back up its dates.Weighted average of how firmly 3 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better
IBM's Hugging Face[1][2][3] blog post 'Granite 4.0 Nano: Just how small can you go?' and the granite-4.0-h-1b and granite-4.0-h-350m model cards give October 28, 2025; It's FOSS reported it the next day.- Start
- Oct 28, 202590% CONFIDENCEsource
Granite 4.0 Nano 1B and 350M Released
IBM releases Granite 4.0 Nano, its smallest models yet: hybrid and plain-transformer versions of roughly 1B and 350M parameters for edge and on-device use, under Apache 2.0.
- On 28 October 2025 IBM released Granite 4.0 Nano, four instruct models and their base counterparts, as its smallest models yet and part of the Granite 4.0 family, designed for edge and on-device applications[1]
- The line-up is Granite 4.0 H 1B (about 1.5B parameters) and H 350M (about 350M), both hybrid state-space designs, plus plain-transformer Granite 4.0 1B and 350M for runtimes such as llama.cpp that did not yet have optimized hybrid support[1]
- They were trained with the same methods and over 15 trillion tokens of data as the original Granite 4.0 models, are Apache 2.0, run natively in vLLM, llama.cpp and MLX, and carry IBM's ISO 42001 certification[1]
- IBM's own comparison against Qwen, LiquidAI, Gemma and others in the 0.2B to 2B range puts the Nano models ahead on its averaged knowledge, math, code and safety benchmarks and on IFEval and BFCLv3 tool calling; these are IBM-run results[1]