When you ship an AI product, you need reliable inference: low latency, clear pricing, autoscaling, and the right model catalog. This category covers model APIs, serverless GPU runners, fine-tuned hosting, and hyperscaler AI platforms that teams actually wire into backends, agents, and creative pipelines—from Replicate and Fal to Vertex, Bedrock, and specialized LPU and GPU clouds.
Browse our complete database of tested and ranked AI applications.
Replicate runs open-source and commercial machine learning models behind a simple HTTP API with per-second billing, webhooks, and autoscaling so you can add image, video, audio, and language inference without owning GPUs.
Fal is a generative media inference platform focused on fast diffusion, video, and audio models with serverless endpoints, queues, and workflows tuned for low-latency production apps.
Together AI provides open-weight and frontier model inference, dedicated endpoints, fine-tuning, and GPU clusters aimed at teams that want open models with serious throughput.
DeepInfra hosts open-weight models behind simple per-token or per-second pricing with autoscaling, aimed at developers who want cheap inference without running their own GPU fleet.
Fireworks AI is a generative inference platform for fast open and proprietary models with serverless deployments, on-demand GPUs, and fine-tuning aimed at production engineering teams.
Modal is a serverless Python platform for running GPUs and CPUs on demand, popular for embedding pipelines, fine-tunes, and custom inference microservices without managing Kubernetes by hand.
RunPod rents GPUs in the cloud with templates for inference, training, and serverless endpoints, aimed at builders who want price-transparent compute.
Baseten helps teams deploy, scale, and monitor custom and open models behind production APIs with autoscaling, observability, and GPU orchestration.
Banana.dev (often paired with Potassium) offers serverless GPU inference for custom models with simple scaling semantics aimed at ML engineers shipping bespoke endpoints.
Hugging Face connects thousands of models to managed inference endpoints and router APIs so teams can serve transformers, diffusion, and embeddings with provider choice behind one integration surface.
OpenRouter is a unified API gateway across many foundation models with per-model pricing, fallbacks, and routing that lets apps switch providers without rewriting client code constantly.
Novita AI provides GPU cloud and serverless APIs for image, video, and LLM workloads with a large marketplace of models aimed at global developers.
Groq offers very fast inference for supported LLMs using its LPU hardware and cloud API, aimed at low-latency assistants, agents, and realtime experiences.
OpenAI's platform API exposes GPT, embedding, image, audio, and realtime models with usage billing, batch endpoints, and fine-tuning for production assistants and agents.
Anthropic's API serves Claude models for text, vision, tool use, and long-context workloads with batching, prompt caching, and enterprise controls for regulated deployments.
Lambda provides cloud GPU instances and clusters widely used for training and high-throughput inference when teams want predictable VMs and networking.
Anyscale builds on Ray for scalable training, batch inference, and online serving patterns used by teams that need custom pipelines beyond a single REST model call.
Amazon Bedrock is AWS's managed service for foundation models from Amazon and partners, with IAM integration, private networking, and governance patterns for enterprise inference.
Microsoft Azure OpenAI Service hosts OpenAI models inside Azure with regional deployment, private networking, and enterprise policy controls for regulated inference workloads.
Vertex AI is Google Cloud's managed ML platform for training, tuning, and serving models—including Gemini and partner models—with enterprise networking, monitoring, and governance.
FriendliAI delivers dedicated and serverless serving for generative models with a focus on efficient GPU utilization and developer-friendly deployment workflows.
SiliconFlow offers high-throughput inference APIs for many open models with competitive pricing, widely used by developers connecting Chinese and global open-weight ecosystems.
Cerebrium is a serverless ML deployment platform for shipping models as scalable APIs with monitoring and versioning—often compared to Modal and Baseten for teams that want fast endpoints without hand-rolling Kubernetes.
Predibase is a low-code platform for fine-tuning and serving open models with declarative configs, aimed at teams shipping specialized models without building a full MLOps department.
Cloudflare Workers AI runs models on Cloudflare's edge network close to users, ideal for lightweight classification, embeddings, and small LLMs inside Workers and Pages backends.
Cerebras offers cloud inference on wafer-scale hardware for selected ultra-large models, targeting extremely high throughput generation for specialized workloads.
SambaNova Cloud delivers DataScale and Reconfigurable Dataflow Unit inference services for enterprises that want full-stack AI hardware and software from one vendor.
Mistral's La Plateforme exposes Mistral family models for chat, embeddings, moderation, and OCR-style workflows with EU-centric deployment options for application backends.
AI21 Labs offers Jurassic and Jamba family language APIs with long-context and structured workflows for enterprise text automation and retrieval-heavy applications.
CoreWeave is a specialized cloud built for AI workloads, offering large-scale GPU clusters and inference infrastructure used by labs and enterprises training and serving big models.
NVIDIA NIM provides optimized inference microservices for popular models on NVIDIA GPUs, designed to drop into Kubernetes and enterprise AI platforms with standardized containers.
Databricks Model Serving deploys ML and generative models next to lakehouse data with unified governance, monitoring, and batch plus realtime patterns inside the Databricks platform.
Snowflake Cortex brings LLM and embedding functions inside the Snowflake SQL environment so teams can run inference co-located with governed enterprise data.
Google AI Studio provides browser and API access to Gemini models, keys, and prototyping tools that feed into Vertex for teams moving from experiment to production.
Perplexity's Sonar API family provides grounded web search and chat completions for apps that need citations, retrieval, and fresh information alongside generation.
Higher parameter counts (70B+) directly correlate with better logic and memory persistence in AI Inference.
Prioritize platforms with End-to-End Encryption or strict "No-Log" policies for sensitive creative sessions.
Our research team monitors API updates and model releases daily to ensure these technical insights remain accurate.
The study and development of new AI technologies and methodologies.
AI-powered search engines and tools for information retrieval.
Freely available AI technologies and platforms that encourage collaboration and innovation.
AI tools to help with programming, code generation, and software development.
Tool-using AI that runs multi-step workflows across browsers, IDEs, SaaS APIs, and messaging—with memory, approvals, and tracing.
View this category list →Found a useful AI tool? Save this directory or share it with your network to help others discover the future of AI.