InceptionMercury
Mercury Edit 2
Mercury Edit 2 uses recent edits and code context to suggest a next change. Inception trained it with human acceptance and rejection feedback.
What the release provides
The training used Kahneman-Tversky Optimization to make suggestions more selective. The API supports fill-in-the-middle completion and next-edit requests. It is also integrated with Zed and ProxyAI.Inception: Introducing Mercury Edit 2
How to use it well
Practical guidance. Measure whether developers accept edits and whether accepted edits remain useful after review. Low response latency alone can hide distracting or oversized suggestions.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API |
| Model identifiers | mercury-edit-2 |
| Context window | 32,000 tokens Advertised 32K per endpoint. Inception Platform model specifications |
| Maximum output | 8,192 tokens API output limit. Inception Platform model specifications |
| Input price | 0.25 USD per million tokens Standard API rate. Inception: Introducing Mercury Edit 2 |
| Output price | 0.75 USD per million tokens Standard API rate. Inception: Introducing Mercury Edit 2 |
| Endpoints | /v1/fim/completions and /v1/edit/completionsInception Platform model specifications |