IBM ships Granite 4.1: dense 8B that matches 32B MoE, plus vision/speech/safety variants — all Apache 2.0
Granite 4.1 covers 3B / 8B / 30B language models, Granite Vision 4.1 (top score on 7 chart/table/KVP extraction benchmarks), two ASR speech models, embeddings, and a Granite Guardian 4.1 safety classifier — every variant under Apache 2.0. The 8B dense model reportedly matches or beats 32B MoE systems.
IBM's April 29 release is among the broader open-weights launches of the year — not a single model but a family of enterprise-targeted components. The 8B parameter dense language model is the headline performance claim: matching or outperforming 32B MoE systems on the critical benchmarks IBM tracks. Multilingual support spans English, German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, Dutch, and Chinese.
Adjacent releases shipping the same day: Granite Vision 4.1 targets enterprise document data extraction (chart, table, key-value-pair) — IBM claims the highest average score across seven benchmarks in this domain. Granite Speech 4.1 2B (autoregressive ASR with translation) and 4.1 2B-NAR (non-autoregressive for fast inference edits) — both ~2B, both Apache 2.0. Granite Guardian 4.1 — safety classifier trained on IBM's AI Risk Atlas for jailbreak / profanity / hallucination detection.
The packaging is what's notable: enterprise-shaped from the start. Most open-weights releases ship the model and leave safety, OCR, ASR as separate problems. IBM is shipping the whole stack.