Free AI Tools

DBRX is an open-source mixture-of-experts LLM from Databricks. The model performs well on code, math, and general reasoning benchmarks. It is available for free use on the Databricks platform and on Hugging Face. Enterprises use DBRX for fine-tuning on private data workloads.

Falcon is an open-source language model family from the Technology Innovation Institute in the UAE. The models are trained on large, high-quality multilingual datasets. Falcon is popular in regions that value independent AI options. Researchers can use the weights under permissive licenses for many applications.

LLaMA is Meta's family of open-weight large language models. The models show strong performance across reasoning, coding, and conversation tasks. Researchers and companies can download the weights and run them on their own hardware. Many open-source AI projects build on LLaMA as their base model.

H2O LLM Studio is an open-source framework for no-code fine-tuning of large language models. Teams upload private data and customize models through a web UI. The framework supports many open base models and training methods. It is aimed at enterprises that want custom LLMs without a deep ML team.

Mistral AI publishes a family of open-weight language models for public use. The lineup includes Mistral 7B, Mixtral mixture-of-experts, and the compact Ministral series. The models are released under permissive licenses for research and production. Developers around the world use Mistral as a strong open alternative to closed models.

MPT is MosaicML's open-source LLM family including MPT-7B and MPT-30B. The models were released under permissive licenses for commercial use. MosaicML later joined Databricks, bringing the models with them. Developers still use MPT for fine-tuning and research projects.

Nemotron is Nvidia's family of open foundation models for enterprise AI. The models are tuned for RAG, agents, and reasoning tasks in business settings. Nvidia publishes them on Hugging Face and in its NIM catalog. Large companies deploy Nemotron on Nvidia GPU systems for production work.

Nous Hermes is Nous Research's family of open-source chat and instruction models. The models build on top of LLaMA and Mistral base weights. The team is known for careful fine-tuning and strong community engagement. Developers pick Hermes when they want a stable, tested open chat model.

OLMo is a fully open language model family from the Allen Institute for AI. It ships with open weights, training data, and the exact training recipes. Researchers use OLMo to study how language models learn. It is one of the most transparent options for serious open LLM research.

OpenLLaMA is an open reproduction of Meta's LLaMA model. It was trained on the RedPajama dataset with permissive licensing. The project lets developers use LLaMA-style models without Meta's license. OpenLLaMA is commonly used as a base for open-source fine-tuning experiments.

Phi is Microsoft's family of small language models with strong reasoning ability. The models are tiny compared to flagship LLMs but achieve surprisingly good scores. Phi is great for running on laptops, phones, and edge devices. Developers use it when speed and cost matter more than sheer size.

Pythia is EleutherAI's research-oriented open LLM family. It was trained on The Pile dataset and released across many model sizes. The project is designed for reproducible studies of how LLMs scale. It is used widely in academic AI interpretability research.

SGLang is an open-source LLM serving framework for fast, structured generation. It is designed for complex program-like workflows that call LLMs many times. Benchmarks show strong throughput compared with other open servers. Researchers and companies use SGLang to serve RAG and agent apps.

Solar LLM is a compact open-source language model from Upstage. It is designed for strong reasoning while running efficiently on modest hardware. The models come in sizes that fit single-GPU setups. Korean enterprises and developers use Solar for on-prem AI work.

Vicuna is an open-source chat-tuned language model from LMSYS. It was built by fine-tuning LLaMA on shared ChatGPT conversations. Early benchmarks showed Vicuna approaching the quality of commercial chatbots. It helped spark the wave of open chat-tuned models in 2023.

WizardLM is a family of open-source language models tuned with the Evol-Instruct method. The approach produces strong instruction following and coding performance. The models come in many sizes for different hardware setups. WizardLM has been a reference point for open-source fine-tuning research.

Xwin is an open-source LLM research project that fine-tunes LLaMA-based models. The team focuses on improved alignment and chat quality. Their releases have led open benchmarks at various times. Developers use Xwin as another option when looking for strong open chat models.

Yi is a bilingual open-source language model family from 01.AI. It offers strong performance in both Chinese and English across chat and reasoning. The models are available for free download on Hugging Face. Developers in Asia use Yi as a leading open alternative for local languages.

Canva Charts is part of the Canva design suite for making graphs and charts. AI suggestions pick the right chart type based on the uploaded data. Users can style charts to match brand colors and typography. It is built for marketers and small businesses who already use Canva.

Chart GPT is a free AI chart generator that runs in the browser. Users paste raw data or write a plain-language prompt and get a chart back. Outputs can be edited, downloaded, or shared with a public link. It is aimed at students, writers, and anyone who needs a quick visual.