Granite 3.1 2B Instruct

Granite 3.1 2B Instruct is IBM's smallest production Granite model — 2 billion parameters targeting edge inference, fast classification, and extraction at minimal compute cost. The headline spec is 128K context at 2B scale, which is uncommon: Llama 3.2 3B and Gemma 2 2B both cap lower, giving Granite 3.1 2B a real edge on long-document classification tasks where you'd otherwise need a larger model. IBM designed the Granite 3 series for enterprise compliance and tool-use scenarios rather than open-ended generation. Apache 2.0 license makes it straightforward to ship commercially. Expect competitive per-token rates on watsonx.ai; Together AI coverage is limited but growing.