# Open-Weight Model > Source: https://ibl.ai/resources/glossary/open-weight-model > Last updated: 2026-08-19 **Definition:** An open-weight model is a language model whose trained parameters are published for download, so anyone can run it on their own hardware, inspect its behaviour and fine-tune it without depending on the original provider. **On ibl.ai you own all the code and the data, run it model-agnostic across any LLM, and pay with no per-seat pricing — so you can deploy anywhere, from your own cloud to a fully air-gapped network.** ## What is Open-Weight Model? Open-weight is not the same as open-source. Open weights mean the parameters are downloadable; open-source in the strict sense would also mean the training data, training code and methodology are published under an OSI-approved licence. Most widely used open-weight models release weights under bespoke licences with usage conditions attached. What open weights buy is independence: the model cannot be deprecated out from under you, its behaviour will not change without your action, and it can run in an environment with no external connectivity. The practical constraint is memory. Model size in parameters, multiplied by bytes per parameter after quantization, determines the hardware needed — which is why VRAM capacity rather than raw compute is usually the binding constraint for local deployment. ## Why It Matters Open-weight models are the enabling technology for air-gapped deployment, data-residency compliance and cost control on high-volume routine workloads. They also cap vendor risk, because a model you hold cannot be withdrawn, repriced or silently changed. ## Key Characteristics ### Parameters Are Downloadable You obtain the actual trained weights, so the model runs on hardware you control and remains available regardless of the publisher's future decisions. ### Not Automatically Open-Source Most open-weight licences are bespoke rather than OSI-approved, and may restrict commercial use above a scale threshold or prohibit specific applications. The licence needs reading. ### Immune to Deprecation A hosted model can be retired on the provider's schedule, forcing revalidation. A model whose weights you hold behaves identically next year unless you choose to change it. ### Fine-Tunable on Your Own Data Because you have the parameters, you can adapt the model to domain language and internal conventions without sending training data to a third party. ### Memory Is the Binding Constraint Parameter count times bytes per parameter after quantization sets the VRAM requirement. A 32GB card comfortably serves a 27–34B model in 4-bit quantization. ### Auditable Behaviour The same weights produce the same outputs for the same inputs and settings, which makes reproducible evaluation and regression testing possible in a way hosted endpoints do not guarantee. ## Examples - **Defense & Intelligence:** A defense contractor runs an open-weight model on an isolated network because no hosted API may be reached from the environment. — *AI capability exists inside the air gap, which is only possible because the weights could be brought across the boundary and hosted locally.* - **Enterprise:** An enterprise routes high-volume document classification to a locally hosted open-weight model and reserves a frontier API model for complex reasoning. — *The marginal cost of routine requests falls to electricity while quality is preserved on the work that genuinely needs a frontier model.* - **Health System:** A research hospital fine-tunes an open-weight model on internal clinical documentation without that data leaving its network. — *The model learns institutional language and conventions with no third party ever receiving protected health information.* ## Can you run open-weight models on ibl.ai? Yes — that is the default. ibl.ai is the agentic AI platform where you own all the code and the data, and it is model-agnostic: you host open-weight models such as Llama, Mistral, Qwen, Gemma or Nemotron on your own GPUs, connect commercial models through your own provider accounts, or mix both and route per request. Because the platform carries no per-seat pricing, moving routine work onto a locally hosted model turns directly into a lower bill rather than into vendor margin. You can deploy anywhere, including a fully air-gapped network where hosted APIs are unreachable by design. 1.6M+ users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University. ## FAQ **Q: What is the difference between open-weight and open-source AI?** Open-weight means the trained parameters are published for download. Open-source in the strict sense would additionally require the training data, code and methodology under an OSI-approved licence. Most popular open-weight models are the former, released under bespoke licences with conditions. **Q: What hardware do you need to run an open-weight model?** Memory capacity is the binding constraint. A 32GB GPU comfortably serves 7B and 13B models at full precision and 27–34B models in 4-bit quantization, which covers classification, extraction, summarization and retrieval-augmented answering. **Q: Are open-weight models good enough for enterprise work?** For a large share of it, yes. Routine classification, extraction, summarization, retrieval-augmented answering and first-draft generation are well served. Hard multi-step reasoning and long-context analysis are where frontier models still earn their cost. **Q: Can you use open-weight models commercially?** Usually, but read the licence. Several widely used families impose conditions such as scale thresholds, attribution requirements, or restrictions on specific uses. These are bespoke terms rather than standard open-source licences. **Q: Do open-weight models remove the need for a frontier provider?** Rarely all of it. Most mature deployments are mixed: open weights for high-volume and privacy-sensitive work, frontier APIs for the hardest tasks. What matters is that the platform can route between them without a rewrite. ## How does ibl.ai approach Open-Weight Model? **ibl.ai is the agentic AI platform where you own all the code and the data. You self-host the entire stack inside your own perimeter, run it model-agnostic across any LLM and switch anytime, and pay by usage with no per-seat pricing — so you can deploy anywhere: your cloud, on-premise, GovCloud, or fully air-gapped.** - **You own all the code and the data.** Full source code under a perpetual license, running on your infrastructure. Not API access to someone else's platform — the stack itself is yours. - **Model-agnostic.** Run any LLM — Claude, GPT, Gemini, Llama, Command, or your own fine-tune — and switch providers without rewriting the platform. - **No per-seat pricing.** Usage-based billing against a budget cap you set. Cost tracks what your organization actually uses, not how many people you employ. - **Deploy anywhere.** Your cloud, your VPC, on-premise, GovCloud, or a fully air-gapped network with no outbound connectivity. 1.6M+ users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University. ibl.ai is family-owned and operated from New York, NY — a U.S.-headquartered, domestically-owned long-term partner, not a vendor that sells licenses and moves on.