Best Free LLM Gateways of 2026

LiteLLM is the Apache 2 / MIT-licensed OSS proxy pick and the largest community among self-hosted LLM gateway proxies. Founded 2023 by BerriAI Inc (Y Combinator W23) in San Francisco. The wedge for free readers: Apache 2 / MIT permissive license, Python proxy that translates 100-plus provider APIs to the OpenAI Chat Completions schema, and self-host with zero licensing cost. Run it as a self-hosted service in front of your model providers and your application code targets the OpenAI SDK only.

Open Source is Apache 2 / MIT-licensed and free forever for self-hosted deployment with 100-plus provider integrations and community support. Cloud Free is a hosted tier with limited request volume. Cloud Pro is the upgrade tier at fifty dollars per user monthly with cost tracking, budgets, team management, and virtual API keys. Enterprise covers self-hosted enterprise with SSO and SOC 2.

The trade-off versus OpenRouter is operational tax; LiteLLM self-host requires Python operational maturity to run reliably. The trade-off versus Cloudflare is delivery model; LiteLLM is a proxy you run in front of providers where Cloudflare is bundled with Workers infrastructure. For free Python-stack readers who want full code control and zero markup, LiteLLM is the right call.

Portkey is the production-gateway pick that fully open-sourced under Apache 2.0 on March 24, 2026 and the cleanest free-forever path for production-scale features. Founded 2023 in San Francisco. The wedge: Apache 2.0 self-host ships caching, fallbacks, prompt management, guardrails, RBAC, and observability across 40-plus providers. Cloud handles around one trillion tokens a day across roughly 24,000 organizations.

Open Source Apache 2.0 is free forever for self-host with Universal API, retries, routing, guardrails, and load balancing. Developer Free hosted covers 10,000 logs a month with key management. Production is the upgrade tier at forty-nine dollars monthly with 100,000 logs and semantic caching. Enterprise covers VPC hosting, SSO, SOC 2 Type II, GDPR, and HIPAA.

The trade-off versus LiteLLM is scope; Portkey is a full production gateway where LiteLLM is a focused proxy. The trade-off versus Cloudflare is delivery model; Portkey is self-hosted rather than bundled with edge. For free production teams who want caching and guardrails on a self-hosted Apache 2.0 gateway, Portkey is the right call.

Cloudflare AI Gateway is the edge-native pick and the cleanest free-forever path for indie developers running serverless apps. Launched 2023 by Cloudflare (founded 2009 in San Francisco). The wedge for free readers: the gateway is bundled with Cloudflare Workers across 300-plus data centers, calls run from the closest Cloudflare PoP, and cached responses serve at the edge with zero infrastructure setup.

Free covers 10,000 requests a day with caching, analytics, rate limits, and OpenAI plus Anthropic plus Workers AI provider support. Workers Paid is the cheapest paid upgrade at five dollars monthly. Workers Standard at the next upgrade tier bundles 50 million Workers requests. Most free indie developers stay on Free indefinitely; Workers Paid is the upgrade when daily volume passes 10,000 requests.

The trade-off versus OpenRouter is provider count; Cloudflare ships fewer model providers than OpenRouter's 60-plus. The trade-off versus LiteLLM is gateway feature breadth; Cloudflare ships caching but lighter prompt management. For free indie readers building edge-native serverless apps, Cloudflare is the right call.

Vercel AI Gateway is the frontend-stack-native pick and the cleanest path for Next.js teams already on Vercel. Launched 2024 by Vercel (founded 2015 in San Francisco). The wedge for free readers: the gateway is bundled with the Vercel Hobby plan at zero cost, ships hundreds of models behind a unified endpoint with zero token markup, and integrates tightly with the AI SDK v5 ecosystem for the cleanest Next.js stack experience.

Hobby is free with the Vercel Hobby plan with hundreds of models, no markup, BYOK support, and built-in observability. Pro is the upgrade tier at twenty dollars per user monthly bundled with Vercel Pro and adds load balancing, fallbacks, budgets, and spend monitoring. Enterprise covers private clusters and SAML SSO. Most free Next.js developers stay on Hobby indefinitely; the Pro upgrade is for production team workflows.

The trade-off versus OpenRouter is provider count; Vercel ships hundreds of models versus OpenRouter's 300-plus. The trade-off versus Cloudflare is edge-native delivery; Vercel runs from Vercel's edge but Cloudflare's edge is broader. For free Next.js readers already on Vercel and using the AI SDK, Vercel AI Gateway is the friction-free path.

OpenRouter is the indie-dev brand reference for LLM gateways and the cleanest path for free model evaluation. Founded 2023 in San Francisco. The wedge for free readers: every account starts with a one-dollar credit that grants full access to all 300-plus models from 60-plus providers behind a single OpenAI-compatible endpoint. No credit card required for evaluation.

Free gives every account a one-dollar starter credit with full access to all models. Pay-as-you-go bills consumed tokens at provider price with no inference markup; a small Stripe processing fee applies on credit purchases. Pro is the upgrade tier at twenty dollars monthly as a credit minimum unlocking higher rate limits and Slack support. Most free indie developers run small evaluations on the starter credit then move to pay-as-you-go.

The trade-off versus Cloudflare is request volume on the free path; OpenRouter's starter credit funds testing rather than ongoing free use. The trade-off versus LiteLLM is OSS optionality; OpenRouter is closed-source SaaS where LiteLLM is Apache 2 / MIT self-host. For free indie readers evaluating multiple models without managing provider keys, OpenRouter is friction-free.