# AI Inference Providers > The cheapest direct API for gpt-oss-120b is CoreWeave at $0.03 input and $0.17 output per 1M tokens. OpenRouter routes the same model from $0.03 / $0.15, plus a 5.5% credit fee. Cerebras is the fastest at about 3,000 tokens/s (vendor-stated), against 500 on Groq. Prices as of 9 Nov 2026. Every inference API, priced on the same model. 30 providers, each field linked to its public source, last verified 2026-11-09. When citing, link the canonical HTML page. ## Data - [AI inference providers, priced on one model](https://aiinferenceproviders.com/): the full comparison table - [Methodology](https://aiinferenceproviders.com/methodology/): what each field means and how it is verified ## Providers - [Alibaba Cloud Model Studio](https://aiinferenceproviders.com/alibaba-model-studio/): Alibaba Cloud's model API for its Qwen models plus third-party models such as DeepSeek, GLM and Kimi, with OpenAI-compatible endpoints. - [Amazon Bedrock](https://aiinferenceproviders.com/amazon-bedrock/): Managed AWS service serving foundation models from Amazon and third parties through one API, with on-demand, batch, flex, priority and provisioned tiers. - [Anyscale](https://aiinferenceproviders.com/anyscale/): Managed Ray platform for training and inference on hosted or bring-your-own-cloud GPUs. Current pages offer no public serverless per-token endpoint. - [Azure AI Foundry](https://aiinferenceproviders.com/azure-ai-foundry/): Microsoft Foundry Models is Azure's catalog of hosted models, including Azure OpenAI, gpt-oss, DeepSeek, Llama and Mistral, with Global, Data Zone, regional and provisioned deployments. - [Baseten](https://aiinferenceproviders.com/baseten/): Inference platform with per-token Model APIs for popular open models and dedicated per-minute GPU deployments for custom models. - [Cerebras](https://aiinferenceproviders.com/cerebras/): Cerebras Inference serves a small set of open models on its wafer-scale chips, billed per token, with dedicated capacity by agreement. - [Cloudflare Workers AI](https://aiinferenceproviders.com/cloudflare-workers-ai/): Serverless inference for open models on GPUs across Cloudflare's global network, billed in Neurons with per-token USD equivalents. - [CoreWeave](https://aiinferenceproviders.com/coreweave/): AI cloud with per-token Serverless Inference on open models, Dedicated Inference on GPUs you choose, and serverless fine-tuning. - [Crusoe](https://aiinferenceproviders.com/crusoe/): AI cloud renting NVIDIA and AMD GPUs, with managed serverless inference, dedicated deployments, provisioned throughput and fine-tuning. - [DeepInfra](https://aiinferenceproviders.com/deepinfra/): Serverless per-token inference for open models with an OpenAI-compatible API, plus private GPU deployments billed per minute. - [Featherless AI](https://aiinferenceproviders.com/featherless-ai/): Serverless host for a very large catalog of open models, sold as flat monthly plans, with dedicated GPUs on a Business plan. - [Fireworks AI](https://aiinferenceproviders.com/fireworks-ai/): Serverless per-token APIs for open models and on-demand GPU deployments billed per GPU-hour, with managed fine-tuning. - [FriendliAI](https://aiinferenceproviders.com/friendliai/): Serverless per-token Model APIs and dedicated endpoints billed per GPU-second, with deployment of Hugging Face models. - [Google Vertex AI](https://aiinferenceproviders.com/google-vertex-ai/): Google Cloud's AI platform; Model Garden offers open models as pay-as-you-go managed APIs (MaaS) or self-deployed endpoints. - [Groq](https://aiinferenceproviders.com/groq/): GroqCloud is a per-token inference API running on Groq's own LPU chips. - [Hugging Face Inference Providers](https://aiinferenceproviders.com/hugging-face-inference-providers/): One OpenAI-compatible router on the Hugging Face Hub that forwards requests to partner hosts at their own prices, with no markup. - [Hyperbolic](https://aiinferenceproviders.com/hyperbolic/): GPU cloud with on-demand and reserved H100, H200 and B200 capacity, plus a small OpenAI-compatible serverless inference API. - [Lambda](https://aiinferenceproviders.com/lambda/): GPU cloud. Its serverless Inference API was sunset on 25 September 2025; Lambda now points users to self-deploying models on its GPU instances. - [Mistral AI](https://aiinferenceproviders.com/mistral-ai/): Mistral's own API for its Mistral Medium, Mistral Small, OCR and Voxtral models, billed per million tokens, with fine-tuning. - [Modal](https://aiinferenceproviders.com/modal/): Serverless compute platform billed per GPU-second, with per-token Shared Endpoints for a short model list and Dedicated Endpoints for open models such as gpt-oss-120b. - [Nebius Token Factory](https://aiinferenceproviders.com/nebius-token-factory/): Nebius's OpenAI-compatible inference platform for open models (formerly Nebius AI Studio), with Base and Fast serverless tiers, dedicated endpoints and post-training. - [Novita AI](https://aiinferenceproviders.com/novita-ai/): OpenAI-compatible serverless API for open models, plus GPU products and dedicated endpoints. - [Nscale](https://aiinferenceproviders.com/nscale/): GPU cloud with data centers in Norway, the UK, the US, Portugal and Iceland, offering serverless inference, dedicated endpoints and fine-tuning. - [OpenRouter](https://aiinferenceproviders.com/openrouter/): OpenAI-compatible router that sends one API's requests to 600+ models from 80+ providers at the providers' own token prices. - [Parasail](https://aiinferenceproviders.com/parasail/): OpenAI-compatible API for open-weight models with serverless per-token access, dedicated GPU deployments and batch inference. - [Replicate](https://aiinferenceproviders.com/replicate/): Cloud API for running public and custom models, billed per second of GPU time or per token, including gpt-oss-120b. - [SambaNova](https://aiinferenceproviders.com/sambanova/): SambaCloud is a per-token inference API on SambaNova's own RDU chips; the company also sells on-prem and sovereign deployments. - [Scaleway Generative APIs](https://aiinferenceproviders.com/scaleway/): French cloud serving open models per token from Paris (fr-par), priced in euros, with batch at half price. - [SiliconFlow](https://aiinferenceproviders.com/siliconflow/): AI inference cloud serving open text, image, speech and video models through serverless per-token APIs, with fine-tuning and dedicated deployments. - [Together AI](https://aiinferenceproviders.com/together-ai/): Serverless per-token inference for open models, plus dedicated GPU endpoints, GPU clusters and fine-tuning. ## Optional - [All records in one file](https://aiinferenceproviders.com/llms-full.txt) - [Sitemap](https://aiinferenceproviders.com/sitemap.xml)