IBM open sources its Granite code models for code generative tasks, trained on 116 programming languages, with models ranging in size from 3B to 34B parameters
IBM's release establishes Granite as a code-focused open-model family spanning 3B to 34B parameters and 116 programming languages. It extends IBM's prior practice of making machine-learning software freely modifiable, while joining a broader field in which Cerebras released a range of GPT-based models as open source.
The move is an early step in a continuing Granite release cadence: IBM later introduced Granite 3.0 open models for enterprise use and subsequently pushed toward smaller, locally runnable Granite 4.0 Nano models. The arc matters because it connects broad code-model availability with IBM's enterprise-oriented model portfolio.
First-order effects
Developers and organizations can evaluate, adapt, and deploy IBM's Granite code models across a wide parameter range rather than relying solely on closed code-generation services.
IBM gains an open distribution channel for Granite in coding workflows, with the 3B-to-34B range allowing users to match model scale to their own compute and task requirements.
Second-order effects
Other providers of code-generation models face added pressure to differentiate on model quality, deployment economics, tooling, or governance rather than access alone.
Enterprise buyers gain another model family to test in internal development environments, strengthening their ability to compare code-model options and avoid concentrating experimentation with a single vendor.
Third-order effects
If successive Granite releases continue to cover both larger enterprise models and smaller local models, code-model competition may increasingly center on deployability and operational control alongside raw capability.
The pattern supports a market in which model weights are more portable while differentiation shifts to the runtime, integration, and governance layers that organizations use around them.
The trend: Open model families are expanding from single releases into tiered portfolios that give enterprise users more control over where and how AI coding workloads run.
.@IBM and @RedHat announces the open sourcing of the Granite LLM This is a great move. IBM has believed in the potential of open source since 2000 so this is on brand for big blue. [image]
Granite Code released! 🧑🏻💻 @IBM just released a family of 8 new open Code LLMs from 3B to 34B parameters trained on 116 programming languages and released under Apache 2.0. 🔥 Granite 8B outperforms other open LLMs like CodeGemma or Mistral on benchmarks and supposedly supports […
Welcome IBM Granite Code LLMs 🤗 > 34B, 20B, 8B & 3B models > Base & Instruct - Apache 2.0 licensed > Trained on 4.5T tokens using depth upscaling > Covers 116 code languages ;) Data: > Uses The Stack for pre training > Filters out low quality code > Performs and exact and [image]
This is a pretty huge deal at @IBM: We're open-sourcing four code LLMs developed at @IBMResearch that stand up to any comparable state-of-the-art models to make the future of coding easier for everyone https://research.ibm.com/... [image]
This is truly an opportunity for everyone to create YOUR AI: An AI that knows about your business and builds on your experience. I'll be around in the @InstructLab Lounge this week at #RHSummit and I hope to see you there. Let's show the world what we can do with open source AI.
Our IBM Granite Code series models are finally released today. Despite the strong code performance that you should definitely check out, I also want to point out that the math reasoning performance of our 8B models is unexpectedly good. Congrats to all our teammates! [image]
It's been a big week for open source AI at IBM and Red Hat. In addition to open sourcing the Granite Code series of LLMs, Red Hat today announced the InstructLab project, a new platform technology that allows a community of developers to contribute new knowledge and skills to an