AI API Providers

Anthropic builds the Claude family of language models — known for strong reasoning, long context, and a safety-first design philosophy. Models are available via the Anthropic API and aggregators like AWS Bedrock and Google Vertex.

OpenAI created GPT and the modern LLM ecosystem. The GPT-4o and o-series models power ChatGPT, the OpenAI API, and Microsoft Azure OpenAI.

Google's Gemini family pushes the long-context frontier with 1M+ token windows, native multimodality, and aggressive pricing on the Flash tier. Available via Google AI Studio and Vertex AI.

Meta publishes the Llama family of open-weight models. Llama 3.1 ranges from a tiny 8B variant to a 405B frontier model, and is hosted by every major inference provider.

DeepSeek is a Chinese AI lab whose open-weight V3 and R1 models redefined the cost frontier for reasoning and general capability.

xAI builds the Grok family of models, integrated with real-time X (Twitter) data. Available through the xAI API and X's premium tier.

Mistral AI is a French AI lab building open and proprietary models with a focus on European languages, function calling, and code-specialized variants.

Qwen is Alibaba's family of open-weight language and multimodal models. The Qwen 2.5 and QwQ series consistently top open-source benchmarks for coding and math, with strong multilingual support across Asian languages.

Cohere builds enterprise-focused LLMs optimized for retrieval-augmented generation (RAG) and tool use. The Command R series is purpose-built for grounded, citation-accurate answers in document Q&A pipelines.

Amazon's Nova family of generative AI models is natively integrated with AWS Bedrock, offering text, image, and video understanding across Pro, Lite, and Micro tiers — optimized for enterprise AWS workloads.

Microsoft's Phi family of small language models (SLMs) is designed for edge and on-device inference. Phi-4 delivers frontier-class reasoning within a 14B parameter footprint, optimized for STEM and coding tasks.

Zhipu AI (Z.ai) is a Beijing-based lab spun out of Tsinghua University, building the open-weight GLM series. The GLM-4.5, 4.6, 4.7 and GLM-5 models are strong agentic and coding performers with vision (GLM-V) variants, popular for self-hosting at a fraction of frontier API prices.

Moonshot AI is the Chinese lab behind the Kimi family. Kimi K2 is a large open-weight Mixture-of-Experts model tuned for agentic tool use, coding, and very long context — a leading open alternative to closed frontier models at low API cost.

NVIDIA publishes the open Nemotron family — models post-trained for reasoning, agentic workflows, and tool use, including Llama-Nemotron variants. They're designed to run efficiently on NVIDIA hardware and are freely available for self-hosting.

MiniMax is a Shanghai-based lab whose MiniMax-M series pairs very long context windows (up to 1M+ tokens) with strong agentic and coding performance at aggressive prices. The models are open-weight and widely hosted.

Perplexity builds the Sonar family — models with built-in web search that return grounded, citation-backed answers. Sonar, Sonar Pro, and the Sonar Reasoning and Deep Research tiers are the engine behind Perplexity's answer product.

Nous Research builds the open-weight Hermes series — steerable, instruction-tuned fine-tunes of Llama with strong function-calling and minimal refusals. A favourite of the open-source community for agents and self-hosting.

ByteDance (TikTok's parent) ships the Seed family of general-purpose models and UI-TARS, a GUI-agent model that controls computers from screenshots. The Seed models offer long context and competitive pricing across flagship and flash tiers.

Arcee AI builds small, efficient enterprise models — the Virtuoso, Coder, and Trinity families — using model-merging and distillation. They target strong quality-per-dollar for coding, reasoning, and on-prem deployment.

AI21 Labs is an Israeli lab whose Jamba models use a hybrid Mamba-Transformer architecture for efficient very-long-context inference. Jamba targets enterprise document and RAG workloads with 256K-token windows.

Reka AI is a research lab building compact multimodal models — the Reka Flash and Edge series understand text, images, audio, and video. Designed to be efficient enough for on-device and edge deployment.

IBM's open-weight Granite family targets enterprise use — small, efficient, commercially-licensed models with strong governance and transparency. Granite is tuned for coding, RAG, and tool use within IBM watsonx.

Tencent builds the Hunyuan family of models, including Mixture-of-Experts variants. Hunyuan powers Tencent's products and is available open-weight for general chat, reasoning, and multilingual workloads.

Inflection AI builds emotionally-intelligent models behind Pi, its personal AI. The Inflection-3 models — Pi (conversational) and Productivity (instruction-following) — target empathetic, safe enterprise assistants.

Liquid AI, an MIT spin-out, builds Liquid Foundation Models (LFMs) using a non-transformer architecture for high efficiency. LFMs are tiny, fast, and designed for on-device and edge inference with low memory footprints.

The Allen Institute for AI (Ai2) builds OLMo — fully open models that release weights, training data, and code together. OLMo is a reference point for reproducible, transparent open-source AI research.

Baidu builds the ERNIE family, China's longest-running large-model line. ERNIE 4.5 includes large Mixture-of-Experts and vision-language variants, available open-weight and via Baidu's cloud.

Writer builds the Palmyra family of enterprise models tuned for business writing, domain-specific knowledge work, and long-context document tasks. Palmyra X5 offers a 1M-token window for whole-document workflows.

Upstage is a South Korean lab building the Solar family — small, efficient models that punch above their parameter count via depth-up-scaling. Solar Pro targets strong document understanding and multilingual performance at low cost.