Vendor · Language models · Audio models
IBM has 9 models in the directory: 7 language models and 2 audio models.
IBM provides enterprise software, cloud, and consulting services, including watsonx tools for building and managing AI solutions. Its predecessor C-T-R was founded in 1911 and adopted the IBM name in 1924.
In the directory: 7 language models and 2 audio models. 9 of them with open weights that can run locally.
Every model there ever was. Models with a profile lead to every benchmark, price and source.
| Granite 4.2 30B | Aug 25, 2026 | Open weights | n/a | n/a | n/a | n/a | 49 |
| Granite 4.2 3B | Aug 25, 2026 | Open weights | n/a | n/a | 54 | n/a | 49 |
| Granite 4.2 8B | Aug 25, 2026 | Open weights | n/a | n/a | 53 | n/a | 53 |
| Granite 4.1 8B | Apr 29, 2026 | Active | n/a | $0.05 / $0.10 | 58 | 131K | 45 |
| Granite 4.0 H Micro | Oct 2, 2025 | Active | n/a | $0.017 / $0.11 | n/a | 131K | 28 |
| Granite-4.0-H-Small | Oct 2, 2025 | Active | n/a | n/a | n/a | 131K | 35 |
| Granite Code 34B | May 2024 | Active | n/a | n/a | n/a | 8K | 32 |
| Granite Speech 5.0 470M TurboCTC | Aug 25, 2026 | Open weights | n/a | 52 |
| Granite Speech 5.0 470M TurboCTC NC | Aug 25, 2026 | Open weights | n/a | 50 |
Language model · Aug 25, 2026
Granite 4.2 30B receives a second supervised fine-tuning phase specifically for agentic coding. IBM increases the share of code and software-engineering data while retaining part of the original training mix. Like 8B, it then completes the agent block for software, terminal, and search tasks.
Language model · Aug 25, 2026
Granite 4.2 3B follows a shorter training schedule than 8B and 30B in IBM’s new series. The compact version receives reasoning and coding training but skips their additional agent block covering software, terminal, and search tasks.
Language model · Aug 25, 2026
For Granite 4.2 8B, IBM adds agent tasks in real coding, terminal, and search environments to foundational training. This block distinguishes 8B from Granite 4.2 3B. It trains multi-step tool use against the task’s actual outcome.
Audio model · Aug 25, 2026
Granite Speech 5.0 470M TurboCTC is a 470-million-parameter model for English speech recognition. Its Conformer architecture uses non-autoregressive CTC decoding and targets edge devices, laptops, and mobile hardware.
Audio model · Aug 25, 2026
Granite Speech 5.0 470M TurboCTC NC uses the same 470-million-parameter class and CTC-oriented Conformer architecture as the commercially usable Granite checkpoint. It targets English audio-to-text recognition on modest hardware.
Language model · Apr 29, 2026
Granite 4.1 8B remains a dense 8B transformer despite its 131,072-token context. IBM documents fill-in-the-middle and lists twelve supported languages including German, English, French, Japanese, and Chinese.
Language model · Oct 2, 2025
Granite 4.0 H Micro combines only four attention layers with 36 Mamba 2 layers. The 3B open-weight model still offers a 131,072-token context, twelve supported languages, and tool calls.
Language model · Oct 2, 2025
Granite-4.0-H-Small combines a hybrid 32-billion-parameter architecture with about 9 billion active and up to 128K context. IBM documents fill-in-the-middle and lists twelve supported languages, including German. The open weights use Apache 2.0.
Language model · May 2024
Granite Code 34B is IBM’s instruction-tuned code checkpoint for programming tasks, logical reasoning, and coding assistants. Its model card describes fine-tuning on permissive instruction data covering code, mathematics, function calling, and general instructions. With 34 billion parameters and an 8,192-token context, it is a focused code specialist rather than a universal long-context chatbot.
| Vendor | Models | Model types | Best language model | Top 10 | Open weights | New in 2026 |
|---|---|---|---|---|---|---|
| Anthropic | 33 | 33 language models | Claude Opus 5.5 (Rank 1 to 3) | 4 | 0 | 13 |
| OpenAI | 63 | 46 language models, 4 image models, 2 video models, 9 audio models, 2 specialist models | GPT-6 Astra (Rank 4 to 8) | 2 | 4 | 27 |
| 66 | 43 language models, 4 image models, 6 video models, 11 audio models, 2 specialist models | Gemini 4 Carbon (Rank 1 to 5) | 2 | 19 | 26 | |
| Meta | 32 | 29 language models, 1 image model, 1 video model, 1 specialist model | Muse Wonder Full (Rank 6 to 9) | 2 | 26 | 7 |
| DeepSeek | 23 | 23 language models | DeepSeek V4.1 Flash 1017 (Rank 13 to 20) | 0 | 23 | 9 |
| Moonshot AI | 12 | 12 language models | Kimi K3.1 Pro (Rank 10 to 13) | 0 | 10 | 7 |
| xAI | 16 | 12 language models, 2 image models, 2 video models | Grok 4.7 (Rank 18 to 24) | 0 | 1 | 12 |
| IBM | 9 | 7 language models, 2 audio models | not ranked yet | 0 | 9 | 6 |
By release date. Grey: retired models.
6 releases since January 2026, by release date.
Prices per 1M tokens for the current language models, according to IBM or the provider linked in each model profile.
| Current model | Input | Output | Context |
|---|---|---|---|
| Granite 4.0 H Micro | $0.017 | $0.11 | 131K |
| Granite 4.1 8B | $0.05 | $0.10 | 131K |
The directory lists 9 Granite models from IBM: 7 language models and 2 audio models. It counts every model IBM ever released, including retired ones. 9 are currently available, the newest being Granite 4.2 30B, Granite 4.2 3B and Granite 4.2 8B.
Granite 4.2 30B, released on Aug 25, 2026. The timeline below shows every release by date.
Through the API per 1M tokens (input / output): Granite 4.0 H Micro $0.017 / $0.11. The prices of all 2 current language models are in the price table.
Granite is developed by IBM. The company is based in Armonk and was founded in 1911.
Yes, 9 of the 9 models have open weights. Each model profile lists how much memory it needs.
Granite 4.1 8B with 58 tokens per second according to Artificial Analysis. A typical model manages 98.
Up to 131,072 tokens (≈ 200 A4 pages), which 3 of the current models reach, including Granite 4.1 8B. Every model's value is in the model list.
Data as of Nov 1, 2026