AI Agent StoreIBM · Provider documented
Granite Switch 4.1.1 30B
Run chat, classification and speech transcription on one server. Separate built-in components handle general decisions and harmful-content checks, so you can reuse the same deployment for several steps of a workflow.
Released 2026-10-14 · First reviewed 2026-10-17 · Checked 2026-11-03.
We still need comparable independent tests. You can compare the confirmed features and limits below, but we cannot yet give this version an Agent Fit or Practical Value rating. Test it on records with known answers, including cases that should go to a person.
What you can use today
You download the model files and supply the hardware and software to run them. Capacity is the amount of information a request can contain; settings and available memory can limit what fits.
Exact downloaded model ID: ibm-granite/granite-switch-4.1-30b. Revision 4.1.1 updates the decision adapter and speech recognizer while keeping the base language model unchanged. Pin the package revision when comparing results; improvements are publisher reports, not new independent measurements.
Model cards date these composed checkpoints October 14; IBM’s launch article is October 15. They combine an existing Granite 4.1 base with decision and Guardian adapters plus a speech encoder. These are distinct from the earlier preview bundles. The cards date revision 4.1.1 October 27; IBM announced it October 29. Use granite-switch 0.1.1 or newer. The announcement and model cards differ on speech language counts, so confirm the languages in your exact package.
The base model has a 128K context window. The exact packaged maximum output and question count are not confirmed. The documented vLLM probability lookup supports 128 option-token IDs per call; more options require repeated calls with the same prompt.
Output: Chat text, probabilities over supplied options, safety checks and speech transcripts. Compare decision models, limits and input costs ↗
Apache 2.0 permits downloading and running the weights. There is no verified per-token hosted price here; budget for hardware, memory and serving software.
Provider release and specifications ↗
Check availability and plans ↗
More setup and licensing details ↗
Test it before switching
Give each model the same examples, tools and definition of success. Record correct outcomes, response time, retries and total cost. Include the time and hardware needed to run it.
Compare caching and long-context costs ↗