# Tuna.ai > Ultrafast AI optimized for quality and speed We engineer inference to serve open-weight models faster. ## Homepage - Quality-first commitment - Ultrafast, powered by Bluefin - Heterogeneous compute - US routing & ZDR - Pay-as-you-go - Self-serve Team - Enterprise - Custom deployments - FAQ ## Company - Company mission - Leadership - Build Tuna with us in New York City - Tuna at a glance ## Account Pages - [Log in](https://tuna.ai/login): Interactive login page. No Markdown version is available. - [Sign up](https://tuna.ai/signup): Interactive signup page. No Markdown version is available. ## Documentation > Tuna's serverless API is OpenAI compatible, with GPU inference and KV caches in US data centers and Zero Data Retention by default. ### Get Started - [Introduction to Tuna](https://tuna.ai/docs.md): Tuna serves open-weight models at ultrafast speeds through an OpenAI compatible API, with Zero Data Retention and US regional routing by default. - [Models & Pricing](https://tuna.ai/docs/models.md): Compare token prices, capabilities, and model IDs for Kimi K3 Ultrafast, DeepSeek V4.1 Flash Ultrafast, and GLM-5.3 Flash Ultrafast on Tuna. - [Quickstart](https://tuna.ai/docs/quickstart.md): Make your first request to the Tuna API with curl, Python, or JavaScript, using an API key and the OpenAI compatible chat completions endpoint. - [Authentication](https://tuna.ai/docs/authentication.md): Authenticate to the Tuna API with a project API key sent as a Bearer token in the Authorization header. - [Tuna Docs for Agents](https://tuna.ai/docs/ai-agents.md): Tuna docs pages are also available as Markdown for AI agents and coding assistants, along with llms.txt and llms-full.txt. ### Security and Privacy - [Privacy and ZDR](https://tuna.ai/docs/privacy-and-zdr.md): Inference requests have Zero Data Retention (ZDR) by default. All users are eligible. Learn about prompts, responses, and caching. - [US Regional Routing](https://tuna.ai/docs/us-regional-routing.md): GPU inference locations under tuna-us/. US regional routing is included in model prices. ### Inference API - [Inference Overview](https://tuna.ai/docs/api/overview.md): Tuna's serverless Inference API runs open-weight models on Tuna's GPUs. See the base URL, endpoints, model ids, and what the API can do. - [Recommended SDKs](https://tuna.ai/docs/sdks.md): Call Tuna with the OpenAI JavaScript SDK, the Vercel AI SDK, or plain HTTP. Compress large request bodies with gzip to upload less. - [Chat Completions](https://tuna.ai/docs/api/chat-completions.md): Generate a model reply from a list of messages with POST /v1/chat/completions, with or without streaming. The API is OpenAI compatible. - [List Model Endpoints](https://tuna.ai/docs/api/models.md): List the models Tuna serves, with their ids and prices, using the public GET /v1/models endpoint. No API key is needed. - [Streaming](https://tuna.ai/docs/api/streaming.md): Stream chat completions as server-sent events. Learn the chunk format, token usage, errors during a stream, and how to parse it yourself. - [Tool Calling](https://tuna.ai/docs/api/tool-calling.md): Models request function calls with names and arguments. Your application validates and runs the calls, then returns results. - [Reasoning Effort](https://tuna.ai/docs/api/reasoning.md): Choose a model's reasoning effort level and read generated reasoning text. Reasoning tokens count toward output usage and cost. - [Image Input](https://tuna.ai/docs/api/image-input.md): Send images to models in a chat completion, as URLs or base64 data. A request can have up to 30 images. - [Input Caching](https://tuna.ai/docs/api/input-caching.md): Tuna reuses the start of repeated prompts, and those cached input tokens cost less. Writing to the cache costs the same as normal input. - [Advanced Model Control](https://tuna.ai/docs/api/model-control.md): Tune replies with max_tokens, temperature, top_p, stop, and the presence and frequency penalties. See what each setting does. - [Structured Outputs (Coming Soon)](https://tuna.ai/docs/api/structured-outputs.md): Structured outputs will make replies match a JSON Schema you provide. They are coming soon. Learn what they are and what to use today. - [Assistant Prefill (Coming Soon)](https://tuna.ai/docs/api/assistant-prefill.md): Assistant prefill will let you write the start of the model's reply, so it continues from there. It is coming soon. - [OpenAI Compatibility](https://tuna.ai/docs/api/openai-compatibility.md): Use the OpenAI SDKs with Tuna's serverless API by changing the base URL and API key. See which OpenAI SDK calls work and how Tuna differs. - [Errors](https://tuna.ai/docs/api/errors.md): The HTTP status codes and error bodies the Tuna API returns when a request fails, and what to do about each one. - [Retries](https://tuna.ai/docs/api/retries.md): Which errors to retry, how long to wait with Retry-After and exponential backoff, and how the SDKs retry for you. - [Rate Limits](https://tuna.ai/docs/api/rate-limits.md): Rate limits are shared by everyone in an organization, and grow with its balance. See what to do about 429 errors. ### Account - [Organizations and Projects](https://tuna.ai/docs/account/organizations.md): How Tuna organizes accounts into organizations and projects, and what the Owner, Admin, Member, and View-only roles can do. - [Billing](https://tuna.ai/docs/account/billing.md): Tuna is prepaid. Add credit to your organization, and each request is charged for its input, cached input, and output tokens. - [x402 Agentic Payments](https://tuna.ai/docs/account/x402.md): x402 is an open standard for paying over HTTP with stablecoins, made for AI agents. Tuna plans to support it soon. - [Pay with Crypto (Coming Soon)](https://tuna.ai/docs/account/crypto.md): Tuna plans to accept crypto payments through a crypto payment partner. They aren't available yet. - [Enterprise](https://tuna.ai/docs/account/enterprise.md): Tuna works with companies on SSO with SAML, billing in arrears, higher rate limits, and features they need, like an Anthropic compatible API.