IBM watsonx.ai

Enterprise ML and AI platform covering PAI-Studio visual workflow builder, PAI-DSW Jupyter notebooks, PAI-EAS elastic inference serving, PAI-Blade inference optimisation, and integration with Alibaba's Qwen foundation models

Alibaba's flagship open-source foundation model family covering Qwen (text), Qwen-VL (vision-language), Qwen-Audio, and Qwen-Coder — accessible via the DashScope API with chat, completion, embeddings, and function-calling endpoints

Next-generation SageMaker (rebranded SageMaker AI) unifying data, analytics, and AI in one workspace — Studio notebooks, HyperPod for foundation-model training at scale, Lakehouse with QuickSight + S3 Tables integration, AutoPilot AutoML, managed training jobs, hosted inference endpoints, and Feature Store, with re:Invent 2024 introducing the unified SageMaker AI workspace and 2025 Summit additions extending it with lakehouse auto-onboarding

Build generative AI applications with foundation models from Anthropic (Claude Opus 4.7 from April 2026), Cohere, Meta, Mistral, Stability AI, TwelveLabs (video understanding), and Amazon's own Nova family — accessed via a single API with fine-tuning, knowledge bases, agents, and a model marketplace for discovery and easy onboarding

AWS-built foundation model family covering text (Micro, Lite, Pro, Premier), image generation (Canvas), and video generation (Reel) — accessed through the Bedrock runtime with tight pricing and low-latency streaming, launched at re:Invent 2024

Enterprise access to OpenAI models including GPT-4, GPT-3.5, and DALL-E with Azure security, private networking, regional deployments, and pay-as-you-go or provisioned throughput

End-to-end platform for building and deploying ML models with automated ML, designer (drag-and-drop), managed compute clusters, MLflow tracking, and responsible AI dashboards

Serverless GPU-backed AI inference at the edge running a catalogue of open-source text, image, speech, and embedding models (Llama, Mistral, Stable Diffusion, Whisper, BGE) with pay-per-neurone pricing and direct binding from Workers code

End-to-end AI platform (formerly MLflow + Mosaic ML) for training, fine-tuning, deploying, and monitoring foundation models and custom ML models on the Lakehouse

AI inference runtime that deploys models to Gcore's edge POPs and routes requests to the nearest GPU-backed endpoint, with support for open-source LLMs and custom model containers

Unified platform to build, deploy, and scale ML models with AutoML, custom training on TPUs and GPUs, model registry, pipelines, feature store, and generative AI studio

Direct API access to Google's most capable multimodal AI models with text, image, audio, and video understanding, long context windows, and function calling support

End-to-end AI development platform with AutoML, data labelling, distributed training on Ascend and GPU clusters, and one-click deployment to cloud or edge

Kakao's Korean-first foundation-model family (Kanana Flash / Essence / Nano) for chat, code, and embeddings — multilingual but tuned for Korean conversational performance

Naver's HyperCLOVA X foundation-model platform for Korean-language LLM workloads — chat completion, embeddings, function calling, RAG over Korean text with strong native-language performance

Managed MLOps platform (formerly Open Data Hub) for training, serving, and monitoring ML models on OpenShift with JupyterHub, KServe, Kubeflow, and PyTorch operators

Fully managed service offering Cohere and Llama large language models

Fully managed machine learning platform with JupyterLab notebooks, conda-environment library, job orchestration, model deployments as HTTPS endpoints, feature store, and model catalog — integrated with Autonomous Database and Object Storage for end-to-end ML workflows

Managed inference for open-source LLMs (Llama, Mistral, DeepSeek) hosted in EU datacentres

Fully managed AI and ML service offering hosted LLMs, vector search, and ML functions inside Snowflake SQL

Tencent's in-house family of large language models (Hunyuan-Pro, Standard, Lite, plus multimodal Hunyuan-Vision) accessible via the Hunyuan API, with enterprise-grade context windows up to 256K, function calling, embeddings, and tuning