Disclosure: I run UberLLM, one of the options below. Every other fact was read on that company's own page on November 20, 2026; links at the bottom.
OpenRouter is the default answer for "one key, many models": hundreds of models behind one OpenAI-compatible API. It charges a 5.5% fee ($0.80 minimum) when you buy credits by card, 5% by crypto, and its free models are limited to 20 requests a minute and 50 a day (1,000 a day once you have bought at least 10 credits). You might look elsewhere for three reasons: a lower per-token price, your own provider keys, or more control. Every option here takes OpenAI-style requests, so switching is a new base URL and key.
If you want a lower price on the same models
DeepInfra. Runs models on its own inference infrastructure, with no upfront costs or contracts. Its price list now includes Claude and Gemini models alongside open models like DeepSeek, Qwen and Llama. Watch for: no OpenAI GPT models on its list.
Novita AI. Open models, also available as a Hugging Face inference provider. Watch for: mostly open models; check that yours is listed.
UberLLM. An API marketplace: sellers list capacity for a model and each request goes to a third-party provider, so prices move with the market. No fee on requests; adding credit by card costs 5.5% + $0.50. The same key also covers image, video, music and speech models. Watch for: prices change during the day, and a request can cost more when cheaper sellers are busy (you can cap it with provider.max_price). No free models, fewer models than OpenRouter, and a short track record: it opened in 2026.
If you want routing and controls
Requesty. Routing rules and fallbacks, with budgets per user and API key (team spend controls look like an Enterprise feature). A 5% service fee is added when you top up credits; there is a free plan with 200 requests a day. Roo Code and Kilo Code both list it as a provider. Watch for: model prices plus 5%, so not cheaper per token.
Vercel AI Gateway. A gateway in your Vercel account with no markup and no platform fee on tokens. Watch for: optional extras cost more (regional endpoints add 5% of inference cost), and you pay the providers' list prices, so it is rarely cheaper than OpenRouter per token.
If you already hold provider keys
LiteLLM. Open-source proxy you run yourself, free under the MIT licence. Watch for: you run and update the server, and pay each provider directly.
Cloudflare AI Gateway. Caching, logs and rate limits in front of your providers, in your Cloudflare account. Provider prices pass through with no markup; 5% is added to credits bought through its Unified Billing. Watch for: its OpenAI-style chat completions endpoint is deprecated for single-model calls; Cloudflare points new integrations to its REST API.
Portkey was a common pick here; it is now sold as Palo Alto Networks' Prisma AIRS AI Gateway, with sales-led pricing, so check whether it still fits a small team.
How to choose
- Lowest per-token price on models OpenRouter also sells: compare the exact model on DeepInfra, Novita and UberLLM. It changes by model; no one is cheapest on everything.
- Spend controls and fallbacks, price not the issue: Requesty or Vercel AI Gateway.
- You have keys and want control: LiteLLM (self-hosted) or Cloudflare AI Gateway (hosted).
- Need free models: stay on OpenRouter.
Sources (read 2026-11-20)
- OpenRouter: https://openrouter.ai/docs/faq , https://openrouter.ai/docs/api-reference/limits
- DeepInfra: https://deepinfra.com/pricing
- Novita on Hugging Face: https://huggingface.co/docs/inference-providers/providers/novita
- Requesty: https://requesty.ai/pricing , https://roocodeinc.github.io/Roo-Code/providers/requesty , https://kilo.ai/docs/providers/requesty
- Vercel AI Gateway: https://vercel.com/docs/ai-gateway/pricing
- LiteLLM: https://litellm.ai
- Cloudflare AI Gateway: https://developers.cloudflare.com/ai-gateway/features/unified-billing/ , https://developers.cloudflare.com/ai-gateway/usage/chat-completion/
- Portkey: https://portkey.ai/docs/product/enterprise-offering
- UberLLM: https://uberllm.dev/pricing , https://uberllm.dev/how-it-works
Top comments (0)