AI gateways and model platforms: pricing compared

Gateway or platform	Fee on tokens	Pricing model	BYOK	Models
Independent gateways
HeliconeUnverified Open source LLM observability platform with an OpenAI-compatible AI gateway that logs, caches and fails over requests across providers. BYOK 0% · 100+ models	None	0% markup on credits; plans by requests and seats (Pro $79 a month)	Yes, 0%	100+
Kong AI GatewayUnverified Set of AI plugins on the Kong API gateway that add multi-provider routing, load balancing, semantic caching, guardrails and token rate limiting. BYOK no	None	$100 a month per model proxied (Konnect Plus); no token fee	No	See page
LiteLLMUnverified Open source Python SDK and proxy server that calls 100+ LLM providers in the OpenAI format, with an enterprise tier for SSO, audit logs and support. BYOK 0% · 100+ models	None	Open source, self-hosted with your own keys; enterprise license by quote	Yes, 0%	100+
Neon AI GatewayUnverified LLM gateway built into each Neon Postgres branch: one Neon credential calls OpenAI, Google, xAI and open-weight models through an OpenAI-compatible endpoint, served via Databricks and paid from prepaid credits. BYOK no · 49 models	None	Provider list prices, no markup; prepaid Neon credits	No	49
Vercel AI GatewayUnverified Managed gateway from Vercel that routes model calls across providers with failover, request logs, budgets and BYOK, billed at provider list price. BYOK 0%	None	Provider list prices, no markup or platform fee on tokens	Yes, 0%	See page
Cloudflare AI GatewayUnverified Cloudflare proxy for AI provider APIs that adds analytics, logging, caching, rate limiting, retries and model fallbacks. BYOK yes	5%	Core features free; 5% fee on Unified Billing credits	Yes	See page
RequestyUnverified Hosted LLM router with fallbacks, load balancing, spend limits and analytics, charging 5% on top of model costs. BYOK yes · 600+ models	5%	5% on top of model costs, pay as you go	Yes	600+
Eden AIUnverified Hosted unified API over 500+ AI models with fallbacks and cost monitoring, funded by prepaid credits with a 5.5% platform fee. BYOK yes · 500+ models	5.5%	Provider prices passed through; 5.5% fee on credit purchases	Yes	500+
OpenRouterUnverified Hosted API that routes requests to hundreds of models from many providers through one OpenAI-compatible endpoint and a prepaid credit balance. BYOK 5% · 500+ models	5.5%	Provider prices passed through; 5.5% fee on card credit purchases (5% crypto)	Yes, 5%	500+
Martian GatewayUnverified Hosted model gateway from Martian offering one API compatible with OpenAI and Anthropic formats across 200+ models. BYOK no · 200+ models	Not published	No public pricing	No	200+
Portkey AI gateway with an MIT-licensed open source core and a hosted plan that adds logs, guardrails, caching and prompt management. BYOK yes · 250+ models	$49/mo	Per logged request: free 10k logs a month, Production from $49 a month	Yes	250+
TrueFoundry AI GatewayUnverified Enterprise AI and MCP gateway with routing, fallbacks, guardrails, budgets and observability, deployable in VPC, on-prem or air-gapped. BYOK no · 1,600+ models	$25/mo	Per user and per request ($25 per user a month on Pro)	No	1,600+
Cloud model platforms
AWS BedrockUnverified Managed AWS service that serves foundation models from Amazon and third-party providers through one API, billed on the customer's AWS account. BYOK no · 19 models	List price	Pass-through per 1M tokens at the lab's list price for Global cross-region inference; Geo and in-region inference cost 10% more; billed on your AWS account	No	19
Azure AI FoundryUnverified Microsoft Azure's model catalog and inference service, including OpenAI models plus Grok, Llama, Mistral, DeepSeek, Cohere, Phi and others, billed on an Azure subscription. BYOK no · 2 models	List price	Per 1M (or 1K) tokens on your Azure bill; Global deployment at OpenAI list price, Data Zone and Regional deployments cost 10% more; PTUs for provisioned throughput	No	2
Google Vertex AIUnverified Google Cloud's managed generative AI service serving Gemini plus partner (Anthropic Claude) and open models (Llama, Mistral, gpt-oss, others) per token. BYOK no · 2 models	List price	Per 1M tokens on your Google Cloud bill; partner models at the lab's list price on the Global endpoint, regional/multi-region endpoints about 10% more	No	2
IBM watsonx.ai IBM's AI studio and inference platform serving IBM Granite and third-party models (Meta, Mistral, OpenAI gpt-oss, others) with pay-as-you-go per-token pricing. BYOK no	Not published	Per 1M tokens (billed as Resource Units, 1 RU = 1,000 tokens incl. input and output); Essentials plan from USD 0/month, Standard from USD 1110/month	No	See page
OCI Generative AIUnverified Oracle Cloud Infrastructure's managed generative AI service offering Cohere, Meta Llama, Google Gemini, xAI Grok and OpenAI gpt-oss models on demand or on dedicated AI clusters. BYOK no · 3 models	Not published	Mixed: per 1M tokens for Google, xAI and OpenAI gpt-oss models; per 10,000 transactions (1 transaction = 1 character of prompt plus response) for Meta and Cohere; dedicated AI clusters per unit hour	No	3
Data platforms
Databricks Mosaic AIUnverified Databricks' pay-per-token Foundation Model APIs (open and proprietary models) with AI Gateway governance on the Databricks platform. BYOK no	List price	DBUs per 1M tokens (Standard pay per token; Priority and Batch tiers separate)	No	See page
Snowflake Cortex AI Snowflake's in-platform LLM functions and REST API that run Claude, GPT, Gemini, Llama, Mistral and other models on Snowflake-managed compute. BYOK no · 6 models	20%	Snowflake AI Credits per 1M tokens (input and output priced separately)	No	6
BigQuery MLUnverified BigQuery ML remote models that call Vertex AI (Gemini Enterprise Agent Platform) models such as Gemini and Claude from SQL via ML.GENERATE_TEXT. BYOK no	Not published	BigQuery bytes processed (on-demand or editions) plus the underlying Vertex AI model charges billed directly by that service; no separate BigQuery per-token rate for LLM calls	No	See page
Developer hubs
Hugging Face InferenceUnverified Hugging Face's router that sends OpenAI-compatible requests to partner inference providers (Groq, Together, Fireworks, Cerebras, and others) under one HF token and bill. BYOK yes · 200+ models	List price	Pass-through at each provider's rate with no Hugging Face markup; pay as you go on your HF account, or bill directly to the provider with your own key	Yes	200+
Cloudflare Workers AIUnverified Cloudflare's serverless GPU inference for open models on its network, billed in Neurons, with AI Gateway as the companion control layer. BYOK no	Not published	Neurons: $0.011 per 1,000 Neurons, with per-model token prices published as the Neuron equivalent	No	See page
NVIDIA NIMUnverified NVIDIA's API catalog of hosted model endpoints for prototyping plus downloadable NIM inference microservices for self-hosting on NVIDIA GPUs. BYOK no	Not published	Hosted API endpoints free for NVIDIA Developer Program prototyping; production self-hosted NIM requires an NVIDIA AI Enterprise license from $4,500 per GPU per year (about $1 per GPU per hour in the cloud); no per-token price	No	n/a

A tick means the vendor's own pricing page or docs say so; a dash means we did not find it stated. Unverified marks a row with facts not yet checked by hand.

Claude Sonnet 5 through each gateway

Calculator

100M input and 20M output tokens a month: $400.00 direct from the lab.

A dash: not on that gateway's own model list. No published token fee: Martian Gateway, Portkey, TrueFoundry AI Gateway.

Questions

What is an AI gateway?

A proxy between your app and model providers that gives you one API for many models, with fallbacks when a provider fails, caching, logs, rate limits and spend controls.

OpenRouter vs Vercel AI Gateway: which is cheaper?

Vercel AI Gateway states no markup and no platform fee on tokens. OpenRouter passes provider prices through but charges a fee when you buy credits (5.5% by card). With your own keys, OpenRouter's first allowance is free and then 5%.

What is BYOK?

Bring your own key: you give the gateway your own provider API key, pay the provider directly, and the gateway charges nothing or a small fee for routing.

Is Claude or GPT more expensive on Bedrock, Azure or Vertex AI?

Usually not in the default region: AWS Bedrock, Azure AI Foundry and Google Vertex AI mostly charge the lab's own list price, and some regional endpoints add about 10%. Each platform page lists every model it sells with the premium against the direct API.

How do Snowflake Cortex and Databricks price LLM calls?

In their own units: Snowflake bills Cortex AI functions in credits per million tokens and Databricks bills Foundation Model APIs in DBUs per million tokens. Their pages here convert those to dollars and state the credit or DBU price assumed.

Can I self-host an AI gateway?

Yes. LiteLLM, Portkey (gateway core) and Helicone publish open-source gateways you can run yourself; you pay only for hosting.

Ways to buy model access

Claude Sonnet 5 through each gateway

Questions