OpenRouter Alternatives

LiteLLM is the open-source answer to OpenRouter. It comes as a Python SDK and as a proxy server, which its documentation describes as a self-hosted OpenAI-compatible gateway. You call 100+ LLMs through one interface, issue virtual keys to teams or apps, track spend per key and manage it all in an admin UI. The gateway is under the MIT licence, so you can run and modify it freely. The free edition already includes virtual keys, budgets and teams, load balancing, rate limits and LLM guardrails; the Enterprise tier adds SSO, JWT auth, audit logs and dedicated support on the same core, at a price you request from sales.

How the price reads against OpenRouter: there is no platform fee on top of your tokens, because you bring your own provider accounts and pay them directly. On OpenRouter's pay-as-you-go plan, buying $1,000 of credits costs $55 in platform fees at 5.5%; with LiteLLM that $55 becomes the cost of running a small server and the hours to maintain it. The trade-off is operations: you host it, you update it, you watch it. Tip: start with the SDK inside one service, then move to the proxy when a second team needs access, so you only run infrastructure when it pays.

Portkey is a hosted AI gateway with observability built in. Every plan includes a universal API, fallbacks, load balancing, retries, prompt management and routing. The free Developer plan records 10k logs per month, keeps logs for 3 days and metrics for 30 days, and comes with simple caching, deterministic guardrails and community support; Portkey itself says it is for prototyping and testing, not production workloads. Production costs $49 a month for 100k recorded logs, with $9 overages per additional 100k requests, 30-day log retention, LLM and partner guardrails, role-based access control, semantic caching and production support. Portkey adds that Production is not recommended when you need custom security controls or data residency guarantees. Enterprise is custom-priced, with SSO, granular budgets and rate limits, private cloud or VPC hosting and SOC2 Type 2, GDPR and HIPAA compliance. If you would rather run it yourself, Portkey also offers an open-source edition of the gateway, with retries, fallbacks, load balancing, guardrails and a basic dashboard, and no request limit.

How the price reads against OpenRouter: Portkey's fee is a flat subscription tied to logs, not a percentage of your spend, and exceeding the log limit on Developer only stops recording logs, not requests. At $49 a month, Production costs less than OpenRouter's 5.5% fee once you spend more than about $890 a month on credits. You bring your own provider keys. Watch the ceiling: the $9 overage applies up to 3M requests a month, so 1M requests cost $130 ($49 plus nine blocks of 100k), and past 3M you move to Enterprise. Tip: tag requests with metadata per customer from day one, so cost per customer is a filter in Portkey rather than a spreadsheet job later.

Vercel AI Gateway is the closest hosted twin of OpenRouter: one endpoint, many models, credits you top up. The gateway charges no markup and no platform fee on tokens: you pay the provider's list price, pay as you go, from AI Gateway Credits. Every Vercel team gets a free tier with $5 of credit per month, limited to free-tier eligible models and lower rate limits per model. Buying credits moves the team to the paid tier, with all available models, custom limits and bring-your-own-key (BYOK) support, still at zero markup, but the $5 monthly credit stops applying once you buy credits. You can set budgets per team, project, API key or team member.

How the price reads against OpenRouter: on $1,000 of model usage, OpenRouter Standard adds $55 in platform fees, Vercel adds nothing on tokens, and you remain responsible for any payment processing fees. The free tiers differ in kind: OpenRouter Free gives 25+ free models with daily request caps, Vercel gives $5 of credit a month on a subset of models. Two details catch BYOK users: when a request with your own key fails, the gateway retries it with Vercel's credentials and charges your credit balance, and gateway budgets exclude BYOK spend, so set caps at the provider too. Some options cost extra: team-wide zero data retention is $0.10 per 1,000 requests, and regional inference passes through the provider's regional price, which can be higher. The limit is depth: evaluations need another tool, and OpenTelemetry traces go to your own tool through Trace Drains, billed at $0.05 per 1,000 traces on Pro and Enterprise. In n8n, it has its own Vercel AI Gateway Chat Model node. If you are leaving Vercel for other reasons, read our guide to switching from Vercel. Tip: free credits start with your first request, so make that request when you are ready to test, not on sign-up day.

Cloudflare AI Gateway sits between your app and the model providers and adds caching, rate limiting and analytics. Its core features (dashboard analytics, caching and rate limiting) are free on all plans, and so is DLP scanning; guardrails are billed as Workers AI inference. Log storage depends on when your account created its first gateway: newer accounts store logs in Workers Logs and pay Workers Logs rates, while older ones keep the legacy caps of 100,000 logs in total on Workers Free and 10,000,000 logs per gateway on Workers Paid. If you would rather not keep accounts with each provider, Unified Billing lets you buy credits from Cloudflare: inference is billed at the provider's per-token rates with no markup, and a 5% fee applies to credits, so $100 of credit costs $105, paid by credit card.

How the price reads against OpenRouter: with your own provider keys, the gateway itself costs nothing; with Unified Billing, the 5% fee sits just under OpenRouter Standard's 5.5%. Caching is where the savings hide: a repeated prompt served from Cloudflare's cache does not reach the provider. The limit is the catalogue: you work from the providers Cloudflare supports rather than one marketplace of 600+ models. OpenRouter is one of those providers, so you can also put the gateway in front of OpenRouter and add caching and analytics without leaving it. If Cloudflare is a broader question for you, see our guide to switching from Cloudflare. Tip: turn on caching only for deterministic calls (classification, extraction), since a cached creative answer will repeat word for word.

TrueFoundry is an AI gateway built for governance: an LLM gateway and an MCP gateway in one, with a registry for prompts, skills and agents. The Developer plan is free forever for up to 3 users, with 10K requests and tool calls per user per month, 5 model accounts, 5 MCP servers, one budget rule, 7-day log retention and community support on Discord. Pro costs $25 per user per month, for unlimited users, 20K requests and tool calls per user per month, $20 per additional 100K requests, built-in guardrails, budget limiting, 30-day log retention, SSO and custom roles. Enterprise runs on a custom annual commitment, with VPC, on-premise deployment, data residency and an SLA.

How the price reads against OpenRouter: TrueFoundry charges per seat, not per token, so a team of 5 on Pro pays $125 a month whatever it spends on models, as long as it stays within its 100K pooled requests and tool calls. That beats a percentage fee once model spend is high, and costs more than OpenRouter when spend is low. The MCP gateway is the distinctive part: agents reach tools and models through the same governed layer. Tip: count requests, not users, before choosing a plan. Pro's 20K per user is pooled across the team, a tool call counts like a request, and overage is $20 per 100K: TrueFoundry's own example puts 7 users and 1M requests at $355 a month.

Together AI is a different kind of alternative: an inference provider rather than a router. It runs open models itself, with serverless inference billed per million tokens, then provisioned throughput, dedicated endpoints, GPU clusters, fine-tuning and post-training as you grow. Each model has an input, cached input and output price; for example GLM-5.3 costs $1.40 per million input tokens, $0.26 cached and $4.40 per million output tokens, and GLM-5.3-Flash costs $0.15 input and $0.50 output.

How the price reads against OpenRouter: if most of your traffic goes to a handful of open models, buying from the host removes the middle layer and its fee, and gives you a direct relationship for rate limits and dedicated capacity. The limit is breadth: you get the models in Together's catalogue, and for anything else you keep the lab's own API, so many teams pair Together with LiteLLM or a hosted gateway. For no-code workflows, see our Together AI integration for Make. Tip: check the cached-input price for long system prompts; on GLM-5.3 it is about a fifth of the standard input price.