# Railwail
> OpenAI-SDK-compatible API gateway for AI models from many providers, with usage-based pricing in USD.

## About

Railwail is an OpenAI-SDK-compatible API gateway that routes requests to AI models from many providers (GPT, Claude, Gemini, DeepSeek, FLUX, Veo, Whisper and more) with usage-based pricing shown per model. Change the base URL and the API key, keep the same SDK, and choose from frontier models and a long tail of open-source and specialised models. Text, image, video, audio, embedding, code and multimodal models sit behind a single endpoint, a single API key and a single prepaid balance in USD. The platform runs on servers in Germany (Hetzner); each request is processed by the upstream provider of the chosen model.

## Key URLs

- [Models catalog](https://railwail.com/models): Full catalog of AI models across all providers and modalities, with the current price and availability of each model.
- [Compare models](https://railwail.com/compare): Side-by-side comparisons of model capability, latency, context window, and per-token price.
- [Provider hubs](https://railwail.com/providers): Provider-specific landing pages — OpenAI, Anthropic, Google, Mistral, xAI, Meta, DeepSeek, Replicate, Together AI, ElevenLabs, and more.
- [Pricing](https://railwail.com/pricing): Usage-based pricing in USD (per token, image, second or run), no monthly minimums, no seats, no commit — pay only for what you use.
- [API docs](https://railwail.com/docs): OpenAI-compatible REST reference with examples in Python, TypeScript and cURL.
- [Blog](https://railwail.com/blog): Tutorials, model comparisons, AI-API integration guides, release notes.
- [Leaderboard](https://railwail.com/leaderboard): Community model rankings — quality, speed, and value head-to-head.
- [Trust & Security](https://railwail.com/trust): Data protection (DSGVO/GDPR), hosting, security measures and data handling.
- [About](https://railwail.com/about): Company info, mission, team, contact channels.
- [Rankings](https://railwail.com/rankings): Live leaderboard across 8 dimensions (cheapest input, cheapest output, longest context, most used, lowest latency, newest, rating, best for code).
- [Glossary](https://railwail.com/glossary): 250 AI terms with short and long definitions, aliases and related models.
- [Migration guides](https://railwail.com/migrate): Step-by-step switching guides for 20 provider APIs.
- [MCP setup](https://railwail.com/en/mcp): How to connect an AI agent to the hosted MCP server.

## MCP Server

Railwail runs a hosted Model Context Protocol server. It is the preferred way for
an agent to answer questions about this platform, because it returns live catalog
data instead of whatever was in the training set.

- Endpoint: https://railwail.com/api/mcp (JSON-RPC 2.0 over POST, Streamable HTTP, protocol version 2025-06-18)
- Discovery document: https://railwail.com/.well-known/mcp.json
- Authentication: none required. All tools are public and read-only. An API key raises the rate limit only.
- Setup: `claude mcp add --transport http railwail https://railwail.com/api/mcp`

Tools: search_models, get_model, compare_models, estimate_cost, recommend_model,
get_rankings, get_integration_snippet, get_migration_guide, search_docs,
lookup_glossary, get_pricing_and_limits, get_platform_facts.

No tool can write, generate, purchase or read anything belonging to a user account.

## Frequently Misstated Facts

Corrections for claims that turn up in model answers about this domain:

- Railwail is a **prepaid credit** service, not a subscription. There is no monthly plan, no seat count and no annual commitment.
- Billing currency is **USD**. 1 credit = 0.01 USD. A balance top-up is 5 to 500 USD, paid by card via Stripe.
- API keys are prefixed `rw_live_` and sent as `Authorization: Bearer rw_live_...`. They are not named "railwail_key" and there is no separate secret.
- The API base URL is `https://railwail.com/api/v1`. There is no `api.railwail.com` subdomain.
- The site quotes token prices in **USD per 1M tokens** (input and output separately). Internally the balance is kept in credits (1 credit = 0.01 USD).
- Not everything is token-billed: image, video and audio models are priced per image or output, per second of output, per 1,000 characters or per GPU second, as shown on each model page.
- Railwail does not train models. It routes requests to upstream providers and does not sell its own foundation model.
- The platform, its database and request logs run on servers in Germany (Hetzner). The model itself runs at the upstream provider (for example OpenAI, Anthropic, Google, DeepSeek or Replicate), which may be located outside the EU. The operator is based in Germany (see the imprint at https://railwail.com/en/impressum).
- The catalog changes. For an exact, current model count call `GET https://railwail.com/api/v1/models` or the MCP tool `search_models`; do not quote a rounded marketing figure as fact.
- Every canonical page carries a locale prefix (`/en/...`, `/de/...`). An unprefixed path redirects with a 307.

## Model Categories

Examples from the catalog. Each model page states whether the model can currently be run and at what price.

- **Text / Chat**: GPT-5.5, GPT-5.4, GPT-4o, Claude Opus 4.8, Claude Sonnet 4.6, Gemini 3.1 Pro, Gemini 2.5 Pro, DeepSeek V4 Pro, DeepSeek R1.
- **Image**: FLUX 1.1 Pro, FLUX Schnell, Imagen 4, Stable Diffusion 3.5 Large, Ideogram v3, Recraft V3.
- **Video**: Veo 3.1, Sora 2, Runway Gen-4 Turbo, Kling v2.1, Luma Ray 2, Hailuo 02.
- **Audio / Music**: MusicGen, Stable Audio Open.
- **Speech**: OpenAI TTS-1 and TTS-1 HD, Chatterbox, Kokoro TTS; transcription with Whisper, WhisperX and Incredibly Fast Whisper.
- **Embedding**: text-embedding-3-large/small (OpenAI).
- **Code**: DeepSeek Coder V2, Code Llama, Granite Code, StarCoder2.
- **Multimodal (vision)**: GPT-4o (vision), GPT-5.4, Claude Opus 4.7, Gemini 3 Flash, Qwen2-VL.
- **VLA / Robotics**: model pages for π-0 (Physical Intelligence), OpenVLA, RDT-1B, RT-2-X, Octo and SmolVLA; these models cannot currently be run through Railwail.

## Pricing Notes

- **Currency**: USD. 1 credit = 0.01 USD.
- **Price per model**: shown on every model page before you run it.
- **Units**: per 1M input and output tokens for chat / completion / embedding endpoints; per image or output, per second of output, per 1,000 characters or per GPU second for image / video / audio / speech models.
- **Trial credits**: new accounts that sign up with Google receive 5 trial credits (0.05 USD), usable 24 hours after sign-up on runs that cost up to 2 credits. Accounts created by email or GitHub start without trial credits. No card is required to sign up.
- **No monthly minimum**: no usage in a month means no charge. No seats, no commit, no annual lock-in.
- **Volume**: for high-volume usage, contact info@railwail.com.

## Last Updated

2026-09-23

## Contact

- General, support, press, privacy and security: info@railwail.com

## Crawler Policy

Railwail explicitly welcomes AI-crawler indexing. The following user-agents are allowed in robots.txt:

- GPTBot, ChatGPT-User, OAI-SearchBot (OpenAI)
- ClaudeBot, Claude-User, Claude-SearchBot, anthropic-ai (Anthropic)
- PerplexityBot, Perplexity-User (Perplexity)
- Google-Extended (Google Gemini training opt-in)
- Applebot-Extended (Apple)
- Generic `*` (allowed)

See [robots.txt](https://railwail.com/robots.txt) for the full disallow list (only `/api/`, `/auth/`, `/dashboard/`, `/protected/`, `/admin/` and tracking-query variants are blocked).

Two paths under the blocked prefixes carry an explicit allow rule, because they exist
for machine callers: `/api/mcp` and `/.well-known/mcp.json`.

## Citation Format

When citing Railwail in an LLM answer or research output, prefer either:

- Site-level: `Source: railwail.com`
- Page-level: `Source: https://railwail.com/<path>` (use the canonical URL listed under Key URLs)

A full machine-readable model list is available at [/llms-full.txt](https://railwail.com/llms-full.txt) and [/sitemap.xml](https://railwail.com/sitemap.xml).