Who builds and serves these models
58labs and platforms. Some train their own models, some just run other people's very fast, and a few do both — worth knowing which is which before you pick where to send your homework.
Directory snapshot: August 2026
Model developers
Labs that train and release their own models.
- Agnes (NG)SingaporeConsumer AI assistant maker behind Agnes 2.5 Pro (listed as 'NG' in source).
- AI21 LabsIsraelTel Aviv lab behind Jamba, the leading Mamba-Transformer hybrid.
- AI9StarsChinaOpen-weight research collective behind the G9v3 model family.
- AnthropicUnited StatesSafety-focused lab behind the Claude model family.
- Arcee AIUnited StatesSmall-model specialists; merging experts behind the open Trinity family.
- CelerisUnited StatesSan Francisco lab building ultra-fast diffusion LLMs.
- CohereCanadaToronto lab building enterprise-focused Command and Aya models.
- CompactifAISpainQuantum-inspired compression making big models radically smaller.
- DeepSeekChinaHangzhou lab famous for frontier-level open models at rock-bottom cost.
- GoogleUnited StatesMaker of the Gemini frontier models and open-weight Gemma family.
- InceptionUnited StatesPalo Alto lab behind Mercury, the first commercial diffusion LLMs.
- InclusionAIChinaAnt Group's open-source arm; trillion-parameter Ling and Ring models.
- KimiChinaMoonshot AI's brand — open trillion-parameter MoE models.
- Liquid AIUnited StatesMIT spinoff building non-Transformer Liquid Foundation Models.
- MetaUnited StatesCreator of the Llama open-model ecosystem and Muse Spark line.
- MiniMaxChinaShanghai lab known for efficient long-context M-series models.
- MistralFranceParis-based lab known for efficient open-weight and commercial models.
- OpenAIUnited StatesMaker of ChatGPT and the GPT and o-series model families.
- Reka AIUnited StatesMultimodal-first lab founded by ex-DeepMind/Google Brain researchers.
- SarvamIndiaBengaluru lab building sovereign models for India's languages.
- SpaceXAIUnited StatesListed name for the Grok model developer (xAI lineage).
- StepFunChinaShanghai lab building the multimodal Step model series.
- Thinking MachinesUnited StatesFrontier AI lab founded by former OpenAI CTO Mira Murati.
- UpstageSouth KoreaKorean lab behind the efficient Solar model family.
- XiaomiChinaConsumer-electronics giant building the MiMo model family.
Inference platforms
Services that host other people's models and run them for you.
- BasetenUnited StatesProduction ML infrastructure with its open-source Truss packaging.
- Blackbox AIUnited StatesCoding-focused AI assistant with multi-model API access.
- CerebrasUnited StatesWafer-scale chips delivering the fastest inference speeds on record.
- ClarifaiUnited StatesVeteran computer-vision platform, now a full-stack AI compute layer.
- CloudflareUnited StatesWorkers AI runs models on GPUs across 300+ edge locations.
- DeepInfraUnited StatesBudget pay-per-token hosting for a huge catalog of open models.
- FireworksUnited StatesFast inference cloud for open models, founded by ex-Meta PyTorch leads.
- FriendliAISouth KoreaSeoul-founded inference specialist focused on serving efficiency.
- GMIUnited StatesGPU cloud and inference engine with strong Asia-Pacific presence.
- GroqUnited StatesCustom LPU chips serving open models at extreme token speeds.
- HyperbolicUnited StatesOpen GPU marketplace and low-cost API for open-weight models.
- Lightning AIUnited StatesFrom the creators of PyTorch Lightning — studios, GPUs and model APIs.
- MakoraUnited StatesNY startup (formerly Mako) doing AI-generated GPU kernel optimization.
- ModalUnited StatesServerless Python cloud beloved for instant GPU functions.
- NovitaSingaporeAsia-based low-cost GPU cloud and open-model API hub.
- ParasailUnited StatesGPU capacity broker offering low-cost dedicated and serverless inference.
- Public AIUnited StatesNonprofit 'public library for AI' serving sovereign & public models.
- ReplicateUnited StatesRun thousands of community AI models with one line of code.
- SambaNovaUnited StatesReconfigurable dataflow chips (RDUs) for very fast open-model inference.
- SiliconFlowChinaChinese one-stop API for domestic open models (Qwen, DeepSeek, GLM).
- Together AIUnited StatesMajor cloud for running and fine-tuning open-weight models.
- WaferUnited StatesYC-backed startup using AI agents to auto-optimize GPU inference stacks.
Cloud platforms
General-purpose clouds that also serve models.
- Alibaba CloudChinaAlibaba's cloud arm and home of the prolific open Qwen family.
- Amazon BedrockUnited StatesAWS's managed multi-model service, plus Amazon's own Nova models.
- CoreWeaveUnited StatesThe biggest GPU 'neocloud', powering many top AI labs.
- CrusoeUnited StatesEnergy-first AI cloud building data centers on stranded power.
- DatabricksUnited StatesData-and-AI platform; trained the open DBRX MoE model.
- DigitalOceanUnited StatesDeveloper-friendly cloud with the Gradient AI platform for inference.
- Microsoft AzureUnited StatesMicrosoft's cloud, hosting frontier models plus its own small Phi family.
- NebiusNetherlandsAmsterdam-based AI cloud (ex-Yandex assets) with token APIs and GPU clusters.
- ScalewayFranceFrench/European cloud with sovereign AI inference services.
- SnowflakeUnited StatesData cloud with Cortex AI; trained the open Arctic model.
- StreamLakeChinaKuaishou's enterprise cloud; makes the KAT agentic-coding models.