Key takeaways
- Open source means the source code is published under a licence that lets anyone use, study, modify and share it.
- Open-weight LLMs such as Llama, Qwen, Mistral, Gemma and DeepSeek can be downloaded and run on your own hardware.
- OpenAI's flagship models and Claude are proprietary; OpenAI has released some open-weight models and tools such as Whisper.
- Ollama, LangChain, Qdrant and pgvector are open source; Pinecone is a proprietary managed service.
- Use open models for privacy, cost control and customisation; use hosted frontier models for the hardest tasks.
Open source means software whose source code is published under a licence that lets anyone use, study, modify and share it. In AI, many "open" language models are more precisely open-weight: you can download and run the trained model, but the training data and full training code are not published. This page explains the meaning, which LLMs are open and which are best, why and how to run them, and answers whether OpenAI, Claude, Ollama, Pinecone and LangChain are open source.
What does open source mean?
Open source means the source code is freely available under a licence that grants everyone the rights to use it for any purpose, study how it works, change it and share copies or improvements. The Open Source Initiative maintains the formal definition.
Common open-source licences include permissive ones such as MIT and Apache 2.0, which allow almost any use including commercial, and copyleft ones such as GPL and AGPL, which require sharing changes under the same licence in certain cases. "Source-available" licences publish code but restrict some uses, such as offering the software as a competing service; those are not open source by the strict definition. Always read the licence before building a product on a project.
What are open-source LLM models?
Open-source or open-weight LLMs are language models whose trained weights are published for download, so you can run, inspect and fine-tune them yourself. Well-known families include Meta's Llama, Alibaba's Qwen, Mistral's open models, Google's Gemma, DeepSeek's models, Microsoft's Phi and OpenAI's open-weight releases.
Also asked as: what are open source llm models
Licences vary: some use permissive licences such as Apache 2.0, while others have custom terms that restrict certain uses or very large companies. Few release full training data, so most are open-weight rather than open source in the strict sense. They are available on Hugging Face and through tools like Ollama, and many come in several sizes and variants, such as instruct, coding and vision models.
What are the best open-source models?
The best open models change every few months. Strong general families include the latest Qwen, Llama, DeepSeek, Mistral and Gemma releases, plus OpenAI's open-weight models; specialised coding and embedding models lead in their niches. The best choice depends on your hardware and task.
Also asked as: best open source models
How to choose:
- Laptop with 8 to 16 GB RAM: 3B to 8B models at 4-bit quantisation.
- Single strong GPU: 14B to 32B models, or mixture-of-experts models with small active size.
- Coding: coder-tuned variants.
- RAG: a good open embedding model plus a mid-size instruct model.
- Languages: test Indian-language performance yourself if you need it.
Check recent independent evaluations and test on your own prompts before settling.
Why use an open-source LLM?
Use open models for privacy, since data never leaves your infrastructure; for cost control at high volume; for customisation through fine-tuning; for offline or on-device use; and for independence from a single vendor's pricing and policy changes.
Also asked as: why use open source llm
Trade-offs to weigh: frontier hosted models are still stronger on the hardest reasoning and coding tasks; you are responsible for hosting, updates, security and evaluation; and GPUs cost money even when idle. A common pattern is to use an open model for routine, high-volume or sensitive tasks and a hosted model for hard ones. FineTune Studio exists because adapting open models with QLoRA is one of the main reasons teams choose them.
How do I run an open-source LLM?
The easiest way is Ollama: install it, then run ollama run llama3.2 or another model, and use the local API from your code. For more control, use llama.cpp or LM Studio; for serving many users on a GPU server, use vLLM.
Also asked as: how to run open source llm
ollama run qwen2.5:7b
import requests
r = requests.post("http://localhost:11434/api/chat", json={
"model": "qwen2.5:7b", "stream": False,
"messages": [{"role": "user", "content": "Explain RAG in two sentences."}]})
print(r.json()["message"]["content"])
Match model size to your memory, check ollama ps to confirm GPU use, and keep the local API private unless protected by authentication.
Is OpenAI open source?
OpenAI's flagship models, such as the GPT models behind ChatGPT, are proprietary and available only through its apps and API. OpenAI has released some open models and tools, including open-weight language models and the Whisper speech recognition model, and open-source libraries such as the Codex CLI.
Despite the name, OpenAI moved away from releasing its main models openly years ago, citing safety and competition. Its open-weight releases are a separate line from its frontier models. So you can run some OpenAI models locally, but not the ones powering ChatGPT. For truly local alternatives to GPT-class assistants, use the strongest open models your hardware allows.
Is Claude an open-source LLM?
No. Claude, made by Anthropic, is a proprietary model family available through claude.ai, Claude Code, the Claude API and cloud platforms such as Amazon Bedrock and Google Cloud Vertex AI. Its weights are not published.
Also asked as: is claude open source llm
Anthropic does publish research, safety work and open standards, such as the Model Context Protocol for connecting AI applications to tools and data, which is open source. If you need a model you can run entirely on your own hardware, choose an open-weight model; if you need Claude's capabilities, use it through its official channels and review the data terms for your use case.
Is Ollama open source?
Yes. Ollama is open source under the MIT licence, with its code on GitHub. It is free to use for running open models locally, and it offers optional cloud features with their own plans.
Ollama builds on the open GGML ecosystem that also powers llama.cpp. The models you run through Ollama have their own licences, separate from Ollama's, so check each model's licence for commercial use. Because it is open source, you can inspect what it does, run it fully offline, and contribute fixes. It is one of the simplest ways to start with open LLMs on Windows, macOS and Linux.
Is Pinecone open source?
No. Pinecone is a proprietary, fully managed vector database service. You use it through its cloud API and client libraries; the database itself is not open source.
If you need an open-source vector database, popular choices include Qdrant, Weaviate, Milvus and Chroma, and pgvector, an open-source extension for PostgreSQL. Open options let you self-host for data control and avoid vendor lock-in, at the cost of running the infrastructure yourself. For many projects, especially student and early-stage ones, pgvector or Chroma is simpler and free, and you can move to a managed service later if scale requires it.
Is LangChain open source?
Yes. LangChain is open source under the MIT licence, with Python and JavaScript libraries on GitHub. LangGraph, its framework for stateful agents, is also open source. The company offers paid hosted products, such as observability and deployment tools, alongside the open libraries.
LlamaIndex, another popular framework for RAG and agents, is also open source under the MIT licence. Open frameworks let you inspect how retrieval and agent loops work, which helps debugging. They change quickly, so pin versions, read migration notes when upgrading, and keep your own code's core logic simple enough that you could replace the framework if needed.
Sources
GenAI engineer and AI product builder with 2+ years shipping production-grade AI systems: RAG pipelines, fine-tuned LLMs, hybrid retrieval and multi-modal apps across vision, speech and OCR, architected end to end from ingestion to deployment. Leads engineering for SCULT INDIA's 14-member team, founded the 500+ member TechVerse Enclave community and has mentored 200+ students. Three hackathon first prizes: Changethon 2025 (IIT Roorkee), Product Genesis at Vividhotsava 2025 (CSJMU) and BYTEBATTLE (MeetKats).
Open to GenAI roles, hackathon judging, mentorship sessions and guest talks at colleges. For campus sessions the inviting institution covers the visit end to end: travel (a First AC train ticket for journeys up to five hours, a flight for anything longer), accommodation, meals, and pickup and drop; the honorarium and workshop fees are negotiable on WhatsApp or on the call, and requests are welcome from institutions that can cover the visit this way. Email pranjulrathour41@gmail.com.
