IBM
IBM ranks #35 of 91 providers on BenchLeader by its best model, with 14 ranked models. Its best model is Granite 4.2 30B at 48.7 on the BenchLeader Index, strongest in knowledge. Prices run from $0.053 to $0.283 per million tokens blended; Granite 4.2 3B is the cheapest ranked option. The fastest is Granite 4.2 3B at 110 tokens per second. 14 of its 21 models ship open weights. It has shipped 6 dated releases, roughly one every 118 days; the last was 25 Aug 2026, so the next would be due between 3 Nov 2026 and 6 Feb 2027 on that pattern.
Granite open models aimed at enterprise deployment through watsonx.
- Provider rank
- #35
- by best model
- Best index
- 48.7
- Granite 4.2 30B
- Ranked models
- 14
- 14 open weights
- Cheapest ranked
- $0.053/M
- Granite 4.2 3B
Release timeline
Each model’s best configuration by release date. The top eight are labelled; hover any point for its name.
- IBM13
Next release
Over the last two years IBM shipped 6 distinct models, one every 118 days on median. The latest arrived on 25 Aug 2026.
If the cadence holds, the next one lands between 3 Nov 2026 and 6 Feb 2027.
An estimate from past release dates only. It knows nothing about roadmaps or announcements.
Fastest ranked model: Granite 4.2 3B at 110 tok/s.
Models
Every IBM model with a result, one row per model; switch to all configurations to see each reasoning effort.
| # | ||||||||
|---|---|---|---|---|---|---|---|---|
| 1 | Granite 4.2 30Bopen ↗ | 48.7±3 | $0.283 | 99 | 21 s | 131k | 2026-08-25 | 2026-10-22 |
| 2 | Granite 4.2 8Bopen ↗ | 46.1±4 | $0.108 | 97 | 21 s | 131k | 2026-08-25 | 2026-10-22 |
| 3 | Granite 4.2 3Bopen ↗ | 44.8±5 | $0.053 | 110 | 19 s | 131k | 2026-08-25 | 2026-10-22 |
| 4 | Granite 3.1 8B | 41.8±2 | – | – | – | – | – | 2026-10-22 |
| 5 | Granite 4.0 H Smallopen ↗ | 41.0±3 | – | – | – | 128k | 2025-09-22 | 2026-10-22 |
| 6 | Granite 4.1 8Bopen ↗ | 40.6±5 | – | – | – | 131k | 2026-04-29 | 2026-10-22 |
| 7 | Granite 4.1 30Bopen ↗ | 39.8±6 | – | – | – | 131k | 2026-04-29 | 2026-04-29 |
| 8 | Granite 4.1 3Bopen ↗ | 39.2±4 | – | – | – | 131k | 2026-04-29 | 2026-04-29 |
| 9 | Granite 3.3 8Bno reasoningopen ↗ | 39.2±5 | $0.085 | 38 | 12 s | 128k | 2025-04-16 | 2025-04-16 · stale |
| 10 | Granite 4.0 H 1Bopen | 38.5±4 | – | – | – | 128k | 2025-10-28 | 2025-10-28 · stale |
| 11 | Granite 4.0 H 350Mopen ↗ | 36.8±5 | – | – | – | 33k | 2025-10-28 | 2025-10-28 · stale |
| 12 | granite-4.0-microopen ↗ | 36.2±5 | – | – | – | 128k | 2025-10-02 | 2025-11-24 · stale |
| 13 | Granite 4.0 Nano 1Bopen | 34.4±5 | – | – | – | 128k | 2025-10-28 | 2025-10-28 · stale |
| 14 | granite-4.0-350mopen ↗ | 33.2±5 | – | – | – | 33k | 2025-10-28 | 2025-10-28 · stale |
| 15 | Granite 3.0 2B | – | – | – | – | – | – | 2026-10-22 |
| 16 | Granite 3.0 8B | – | – | – | – | – | – | 2026-10-22 |
| 17 | Granite 3.1 2B | – | – | – | – | – | – | 2026-10-22 |
| 18 | Granite 4.0 Small | – | – | – | – | – | – | 2025-11-24 · stale |
| 19 | Granite 20B Functioncalling | – | – | – | – | – | – | – |
| 20 | Granite 3.2 8B | – | – | – | – | – | – | – |
| 21 | Granite 4.0 Microopen ↗ | – | $0.041 | 34 | 0.49 s | 131k | – | – |
Latest news
- IBM CEO Says Open AI Models Are 90% as Good as Frontier Models: ‘Why Pay 10x More?’Benzinga · 22 Oct 2026
- NASA, IBM launch new AI model for studying the moonSpace · 12 Sept 2026
- The new-hire test for AI decision modelsIBM · 20 Oct 2026
- IBM-backed llm-d framework adds controls for running AI models in-houseSiliconANGLE · 20 Oct 2026
- NASA, IBM Launch AI Foundation Model for Lunar ScienceNASA Science (.gov) · 10 Sept 2026
- New course teaches how AI models run on modern acceleratorsIBM Research · 19 Oct 2026
- IBM CEO says open weight models are now irresistibly good and cheap on-premThe Register · 22 Oct 2026
- IBM, NASA launch AI model to help map ice, craters on MoonReuters · 10 Sept 2026