Gateway comparison
OpenRouter alternatives: 7 AI API gateways compared
If you are choosing an API gateway to sit between your code and model providers, these are the questions that actually decide it: how you pay, what the gateway takes on top of the provider rate, which protocols it speaks, whether Claude Code or Codex work by changing one base URL, and whether you can run it yourself. The table below answers them for seven gateways and for this site, using each vendor's own published pages on 2026-11-17.
| Gateway | How you pay | Fee on top of provider rate | Model coverage | OpenAI Chat / Responses | Anthropic Messages | Claude Code / Codex via base URL | Routing and failover | Self-host | Claude Sonnet 4.6 |
|---|---|---|---|---|---|---|---|---|---|
| OpenRouter | Prepaid credits, deducted per request; Enterprise adds invoicing | None on inference; 5.5% on credit purchases (8% Business), 5% on crypto top-up | 600+ models, 80+ providers: Claude, GPT, Gemini, DeepSeek, Qwen, Kimi, GLM, Grok | Yes, /v1/chat/completions and /v1/responses | Yes, /v1/messages | Yes, documented; some features are only guaranteed on Anthropic first-party providersSetup guide | Automatic provider routing by price, throughput, latency and errors; fallbacks, allowlists, max price | No | $3.00 / $15.00per 1M tokens |
| Requesty | Prepaid USD balance with optional auto top-up; BYOK supported | 5% on every top-up; 0% on BYOK | 600+ models, 30+ providers: Claude, GPT, Gemini, DeepSeek, Qwen, Kimi, GLM, Grok, open-weight | Yes, including a native-shaping responses prefix | Yes, /anthropic/v1/messages | Yes, documented with managed policy model namesSetup guide | Fallback policies, provider pinning, latency routing, prompt caching, budgets and guardrails | No | $3.00 / $15.00per 1M tokens |
| Portkey | BYOK gateway: your own provider keys, subscription plus per-request volume | No markup on tokens; plan fee per request ($49/mo Production, then $9 per 100K) | 250+ providers: Claude, GPT, Gemini, DeepSeek, Qwen, Kimi, GLM, Grok | Yes, OpenAI SDK superset | Yes, Messages endpoint | Yes, documentedSetup guide | Configs with fallbacks, load balancing, retries, conditional routing, semantic caching, MCP gateway | Yes | Not sold; you pay the provider with your own key |
| Together AI | Prepaid credits, $15 minimum first purchase; no BYOK for other providers | None stated; per-token serverless rates listed directly | Open-weight only: DeepSeek, Qwen, Kimi, GLM, Llama, MiniMax, GPT-OSS; no Claude or GPT | Yes, /v1/chat/completions and /v1/responses | Yes, but Together model names only; Claude names return 404 | Yes for Together models; not a way to reach ClaudeSetup guide | Dedicated endpoints, prefix caching, batch API, one virtual model; no cross-provider failover | No | Not sold |
| Vercel AI Gateway | Prepaid AI Gateway credits plus BYOK; enterprise invoicing | Zero markup and zero platform fee on tokens; regional, allowlist, ZDR and BYOK surcharges per 1K requests | 474 listed models from 70+ providers: Claude, GPT, Gemini, DeepSeek, Qwen, Kimi, GLM, Grok | Yes, and Codex uses the responses wire API | Yes, with a dedicated Claude Code endpoint | Yes, documentedSetup guide | Provider failover, model fallbacks, provider ordering, budgets that reject over-limit requests | No | $3.00 / $15.00per 1M tokens |
| Cloudflare AI Gateway | BYOK free, or unified billing with prepaid Cloudflare credits; standard billing at cycle end | No markup on inference; 5% on unified-billing credit purchases | About 24 provider integrations passed through; no unified catalog of its own | Yes, OpenAI SDK compatible | Yes, provider path /anthropic/v1/messages | Only through the Sandbox coding-agent runner; no step-by-step base URL guideSetup guide | Caching, rate limiting, retries, model fallback, dynamic routing, guardrails, DLP | No | Not published; provider rates passed through |
| LiteLLM | Free self-hosted open source, or an annual enterprise licence | No token markup; you pay providers with your own keys. Enterprise priced per deployment, not per token | 100+ providers: Claude, GPT, Gemini, DeepSeek, Qwen, Kimi, GLM, Grok, Bedrock, Vertex | Yes, /v1/chat/completions proxy | Yes, translated, passthrough or raw /anthropic | Yes, documented for the proxySetup guide | Router retries and fallbacks, load balancing, virtual keys with budgets, caching, Prometheus metrics | Yes | Not sold; you pay the provider with your own key |
| Nbility this site | Prepaid credits by card or Alipay in CNY, or USD through Waffo Pancake; no minimum tier | Per-group multiplier on the model's official price; no separate top-up fee | Claude, GPT, Gemini, DeepSeek, Qwen, Kimi, GLM, Grok, MiniMax, Doubao and more; see the Models page for the current list | Yes, Chat Completions and Responses | Yes, native /v1/messages | Yes, documented for Claude Code, Codex and CursorSetup guide | Group-based routing with per-key limits, usage logs and rate limits | No | Priced per group; see the Claude pricing page for the current lowest public group |
Facts checked against each vendor's own pages on 2026-11-17. Prices, model counts and fees change; confirm anything that decides your purchase before you commit.
When this site is the wrong choice
Reasonable fit
- You are an individual developer or a small team paying by card or Alipay and want one endpoint for Claude, GPT, Gemini and Chinese models.
- You run Claude Code, Codex or Cursor and would rather change a base URL than route through another gateway's configuration layer.
- You want the per-group price visible before you send a request, with usage logs you can reconcile after.
Look elsewhere
- You need enterprise procurement, a purchase order and compliant invoicing in your own currency.
- You need to self-host the gateway, keep every request inside your own network, or use a vendor with a signed DPA and regional data residency.
- You need provider-level failover across many upstream accounts with SLA credits; a multi-provider router such as Portkey or LiteLLM fits that better.
Common questions
- Is this site cheaper than OpenRouter?
- Not always, and not for every model. OpenRouter charges the provider list price with a fee on credit purchases; this site applies a per-group multiplier to the official price. For models where the public group multiplier is below the effective OpenRouter rate the request costs less here, and for others it costs more. Compare the per-1M prices on the pricing and vendor pages before switching.
- Can I keep my OpenAI or Anthropic SDK code?
- Yes. Point the SDK's base URL at this site and keep the same request shape. Chat Completions, Responses and Anthropic Messages are all served, so moving from OpenRouter or from the providers directly is a base URL and key change, not a rewrite.
- Can I self-host this gateway?
- No. This site is a hosted service. If self-hosting is a requirement, LiteLLM is the closest option in the table: open source, MIT licensed, and runnable inside your own network.
- Which payment methods work?
- CNY top-ups by card or Alipay through the domestic payment gateway, and USD top-ups through Waffo Pancake, a merchant of record that also handles the tax side. There is no enterprise invoicing, so teams that need purchase orders should use a vendor that provides them.
Check the current prices
Every price on the pricing pages is read from the live gateway, not from a table that was copied once.