How to Use the Mercury 2.5 API on OpenRouter

01

Before you start

OpenRouter exposes Mercury 2.5 Preview under inception/mercury-2.5-preview. You need an OpenRouter API key and should keep it in an environment variable or secret manager rather than hard-coding it.

OpenRouter’s quickstart uses the standard /api/v1/chat/completions endpoint and says catalog model slugs can be substituted into requests.

02

Make a basic Mercury 2.5 request

A minimal Python request uses requests.post("https://openrouter.ai/api/v1/chat/completions", ...) with an Authorization: Bearer <OPENROUTER_API_KEY> header, model: "inception/mercury-2.5-preview", and a standard messages array.

OpenRouter-specific attribution headers are optional. Add production timeouts, retries and error handling around the minimal request, and keep the API key outside source code.

03

Use Mercury 2.5 with the OpenAI SDK

OpenRouter explicitly documents the OpenAI SDK as a drop-in option. Configure base_url="https://openrouter.ai/api/v1", supply your OpenRouter API key, and pass inception/mercury-2.5-preview as the model in client.chat.completions.create(...).

This lets an existing OpenAI-SDK integration test Mercury 2.5 without changing the overall chat-completions code structure.

04

Current Mercury 2.5 API pricing

Inception currently lists standard Mercury 2.5 rates of $0.20/M input, $0.75/M output and $0.02/M cached input. OpenRouter currently applies an 80% Inception discount through September 8, 2026 at 07:00 UTC, showing $0.04/M input, $0.15/M output and $0.004/M cached input.

For example, 100,000 input tokens plus 10,000 output tokens at the promotional rates implies about $0.0055 in token charges. That excludes retries, repeated agent loops and other application behavior.

05

Context, tool calling and structured output

Inception labels Mercury 2.5 with a 256K context window; OpenRouter currently displays 260K and up to 65,536 completion tokens. Use the active provider’s documented limit in production.

OpenRouter documents tools and tool_choice, parallel tool calls and JSON-schema structured output. Validate tool selection and returned structured data before consequential automated actions.

06

Production checklist

1. Pin the exact OpenRouter model slug.
2. Recheck pricing after the September 8 promotion.
3. Set token and timeout limits.
4. Test tool-call and structured-output reliability.
5. Measure end-to-end latency and task success.
6. Define fallback behavior for a Preview model.
7. Validate outputs before external writes.
8. Review provider data-handling requirements for your workload.

Inception’s public direct-API request still demonstrates mercury-2. Until a Mercury 2.5 direct identifier is explicitly documented, do not copy the OpenRouter slug into Inception’s endpoint by assumption.