IBM's Granite line: open-licensed speech models and time-series ensembles
Where it stands: IBM announced Granite Speech 5.0 2B, a 2-billion-parameter multilingual speech-recognition and translation model under Apache 2.0, with a claimed #2 rank on the Open ASR Leaderboard at 5.68% error rate; its weights are not yet confirmed downloadable. IBM Research also published a tutorial on ensemble forecasting with Granite time-series models. Watch leaderboard verification, actual weights and hosting terms, and whether Granite releases continue.
Latest
- 2026-10-22 (~4:00 AM ET, MarkTechPost — marktechpost.com/2026): IBM announced Granite Speech 5.0 2B — a 2-billion-parameter multilingual speech-recognition/translation model under the Apache 2.0 license, with a vendor-cited #2 rank on the Open ASR Leaderboard (5.68% error rate). As of this coverage the model is announced, not confirmed downloadable — weights availability is the thread's first watch. Same day (~3:51 AM ET), IBM Research published a Hugging Face blog on ensemble forecasting with Granite time-series models (huggingface.co/blog) — a technical tutorial suggesting an active open-model cadence. Watch: leaderboard verification, actual weights/hosting terms, whether the Granite line sustains releases.