AI Models

One API key for every available model: compare prices in USD, filter by provider and modality, then try a model in the playground.

Available: 274of 490
Providers: 34
Price range: $0.00030 – $4.80per default run

Quick picks

/

273 models

More models (213)

211 models currently unavailable

Their pages stay online, but they can’t be run at the moment.

All AI models, one API — your unified catalog

Railwail gives you a single OpenAI-compatible endpoint for AI models from many providers — GPT-5.5 and Claude Opus 4.8 for reasoning, Gemini 3.1 Pro for long contexts, FLUX 1.1 Pro for photorealistic images, Veo 3.1 for video with synced audio, Whisper for speech-to-text, OpenAI TTS for voice and OpenAI text-embedding-3 for embeddings. You pick a model, change one parameter in your request, and ship. No new SDK, no new auth flow, no provider lock-in — the catalog above lists every model with its current price in USD and whether it can currently be run.

The pricing is transparent and on-demand: you see the price before you call, you pay per token (or per image, or per second for video and audio), and there are no monthly minimums, no seat fees, and no surprise overage charges. Accounts that sign up with Google get trial credits for the cheaper models, so you can try a few prompts before topping up. Switching between flagships is a one-line change: replace `model: "gpt-5-4"` with `model: "claude-sonnet-4-6"` or `model: "gemini-3-1-pro"` and the rest of your code keeps working. That same surface covers cheap, fast budget tiers like GPT-5.4 Nano, Claude Haiku, Gemini 3 Flash and DeepSeek V4 Flash when latency or unit cost matters more than peak quality.

The platform runs on servers in Germany, Railwail does not train on customer prompts, and every model page names the provider that processes the request, so your compliance team can check each one. Compared with OpenRouter or Together AI, the difference is an operator based in Germany, a prepaid balance in USD and one catalog that covers ten categories — text, image, video, audio, text-to-speech, speech-to-text, embeddings, code, multimodal, and vision-language-action robotics — so a single integration handles your chatbot, your image pipeline, your transcripts, and your RAG retriever without juggling five SDKs.

Typical tasks

Customer support chatbots
Code generation in IDEs
Long-document summarization
Image generation pipelines
Multilingual translation
Voice transcription at scale

Guides by use case

Model comparisons

Frequently asked questions

How do I choose the right AI model for my project?

Start from your workload, not from the leaderboard. For high-volume classification or extraction, pick a fast budget tier (Gemini Flash, GPT-5 Mini, Claude Haiku, DeepSeek V3) — quality is 90% of flagship at 5-10% of the cost. For user-facing reasoning, long-form writing, or code review, pick a flagship (GPT-5, Claude 4.6 Sonnet, Gemini 3 Pro). For images use FLUX 1.1 Pro or Imagen 4; for video Veo 3.1 or Runway Gen-4; for transcripts Whisper or WhisperX. The catalog above can be filtered by category and sorted by price, so you can match the trade-off triangle that matters to your product.

What's the difference between LLMs, image models, and VLA models?

LLMs (large language models) read and write text — GPT-5, Claude 4.6, Gemini 3. Image models like FLUX, Imagen, and Stable Diffusion turn a text prompt into a raster image. Video models like Veo 3 and Kling produce motion clips. Audio models cover music (MusicGen, Stable Audio Open) and sound effects. Speech-to-text turns audio into transcripts; text-to-speech goes the other way. Embedding models compress meaning into vectors for retrieval and search. Code models are LLMs fine-tuned for programming. Multimodal models accept text plus images or video as input. VLA (vision-language-action) models close the loop with motor control — they output robot actions, not text. Railwail lists all of them in one catalog; text, image, video, speech and embedding models run through the same OpenAI-compatible API.

How does Railwail's pricing work?

Pure pay-as-you-go in USD from a prepaid balance (1 credit = 0.01 USD). Text models bill per input and output token at rates shown on every model page; a typical prompt costs fractions of a cent. Image generation bills per image, video per second of output or per clip, speech per 1,000 characters, and some open-weights models by GPU time — the model page lists the exact rate before you call. No monthly minimums, no seat fees, no commitment. The dashboard shows your usage and transaction history, and in the settings you can set a monthly spend limit; new runs are blocked once it is reached.

Do I get free credits to start?

Accounts that sign up with Google receive 10 trial credits (0.10 USD). They can be used 24 hours after sign-up on runs that cost up to 2 credits each, which covers short prompts on the cheaper text models and budget image models such as FLUX Schnell. Accounts created by email or GitHub start without trial credits. No credit card is required to sign up and there is no auto-conversion to paid: when your balance runs out, calls simply stop until you top up (from 5 USD).

How does Railwail compare to OpenRouter or Together AI?

The OpenAI-compatible API surface is similar, so migrating from OpenRouter is mostly a base-URL change. Railwail is operated from Germany, runs its platform on servers in Germany and bills a prepaid balance in USD, with VAT handled at billing time and no monthly minimums or seat fees. Like the other gateways, it forwards each request to the provider of the chosen model, and that provider may process the data outside the EU. Features such as streaming, tool calling and automatic provider failover are not available through Railwail yet — check that before you migrate a workload that depends on them.

Is Railwail GDPR/DSGVO compliant?

Railwail runs its platform, database and request logs on servers in Germany and does not use customer prompts for model training. Each request is processed by the provider of the chosen model; many providers are located outside the EU, and every model page names the provider so your DPO can check it before approving a model. OpenAI and Anthropic state that they do not train on API data by default; other providers' terms differ. Business customers can ask for a data processing agreement, and requests for access or deletion can be sent to [email protected]. Details are on the trust page and in the privacy policy.

Can I switch between models without code changes?

Yes — the whole point of the unified catalog. Railwail exposes an OpenAI-compatible endpoint, so SDKs that talk to OpenAI (openai-python, openai-node, the official SDKs in other languages, plus LangChain, LlamaIndex, Vercel AI SDK and similar) work by changing just the `baseURL` and the API key. After that, swapping `model: "gpt-5-4"` for `model: "claude-sonnet-4-6"` or `model: "gemini-3-1-pro"` is a one-line change; messages, temperature, max_tokens, top_p and stop keep the same shape. `tools`, `response_format` and `stream` are currently not forwarded, and provider-specific extras such as Anthropic's prompt-caching headers are not passed through.

What happens if a provider has downtime?

Each model in the catalog is served by one named upstream provider. If that provider fails, the request returns an error and the credits reserved for it are refunded to your balance. Automatic failover to other providers is not offered as a feature, so build retries or a second model into your own code if a workload must not fail. There is no public status page at the moment.

Explore more

Prices, comparisons and tools in more depth.

Build with one API

Every available model through one OpenAI-compatible API. Prepaid credits in USD, no subscription.

Create account API docs