# Autter Models

> Two hosted security models, autter-large and autter-flash, through the OpenAI and Anthropic SDKs. Prepaid credits, no added latency, nothing stored by default.

Autter Models is the model company by Autter (https://autter.dev). Models that read what attackers write, governed by your rules.
The API is in waitlist mode: access opens in waves (https://models.autter.dev/waitlist). The same models already run inside the hosted Autter platform at https://autter.dev, where every edge case is handled.

## Models

- [Autter Large](https://models.autter.dev/models/autter-large): model id `autter-large`, 131k context, 33k max output, vision, reasoning. $3 in, $0.5 cached, $5 out per 1M tokens. Reads a whole repository or a day of logs in one context, reasons about it, and writes the report.
- [Autter Flash](https://models.autter.dev/models/autter-flash): model id `autter-flash`, 131k context, 16k max output, streaming. $1 in, $0.3 cached, $2 out per 1M tokens. Triages every request, diff hunk and log line as it arrives, and flags what Large should read.

## API

- Base URL for OpenAI SDKs: `https://inference.autter.dev/v1` (Chat Completions, Responses)
- Base URL for Anthropic SDKs: `https://inference.autter.dev` (Messages)
- Auth: `Authorization: Bearer <key>` or `x-api-key: <key>`. Keys are issued from the waitlist in waves.
- [OpenAPI spec](https://models.autter.dev/openapi.yaml)
- Errors: 401 bad key, 402 empty balance, 413 input over the context window, 429 rate limit with retry-after.

## Benchmark (CyberGym)

- autter-large: 88.1% (Autter, open to anyone)
- GPT-5.5-Cyber: 85.6% (OpenAI's dedicated cyber model, gated)
- autter-flash: 84.2% (Autter, open to anyone)
- GPT-5.6-Sol: 83.6% (OpenAI's latest general model)
- DeepSeek-V4-Pro: 83.3% (DeepSeek's top open-weight model)
- Claude Mythos Preview: 83.1% (Anthropic, preview access)

## Pages

- [Home](https://models.autter.dev/): What the models do, the demo transcripts, benchmark, pricing and FAQ.
- [Autter Large](https://models.autter.dev/models/autter-large): Reads a whole repository or a day of logs in one context, reasons about it, and writes the report.
- [Autter Flash](https://models.autter.dev/models/autter-flash): Triages every request, diff hunk and log line as it arrives, and flags what Large should read.
- [The story](https://models.autter.dev/story): 150,000 CVE exploits, each proved in a sandbox, and the model trained on them.
- [Pricing](https://models.autter.dev/pricing): Per-token rates, prepaid credit packs and the billing rules.
- [Docs](https://models.autter.dev/docs): Base URL, endpoint shapes, streaming, errors, credits, data handling, keys, agents.
- [Waitlist](https://models.autter.dev/waitlist): Access opens in waves. The hosted version is live on autter.dev today.

## For agents

- MCP server: `https://models.autter.dev/mcp` (Streamable HTTP, no key). Tools: list_models, get_model, get_pricing, get_benchmark, get_quickstart, search_docs, join_waitlist. [Server Card](https://models.autter.dev/.well-known/mcp/server-card.json)
- [Agent skills](https://models.autter.dev/.well-known/agent-skills/index.json): using the API, joining the waitlist
- [AI catalog](https://models.autter.dev/.well-known/ai-catalog.json) and [API catalog](https://models.autter.dev/.well-known/api-catalog)
- [auth.md](https://models.autter.dev/auth.md): how agents get access

## Optional

- [llms-full.txt](https://models.autter.dev/llms-full.txt): every page as Markdown
- [Autter platform](https://autter.dev): the hosted product the models power
- Contact: hello@autter.dev