Mercury 2
A diffusion language model for low-latency routing, classification, and structured output.
Mercury · Generally available
Overview
Mercury 2 is Inception's reasoning diffusion language model, designed to generate text through parallel refinement instead of conventional token-by-token decoding.
In Manywise, it is used only as a bounded model router when deterministic filtering leaves several plausible answer models.
+Pros
- Low-latency diffusion-based generation
- Native schema-aligned JSON output
- 128K context and tool support
-Cons
- OpenRouter latency and output behavior still require deterministic timeouts and validation
- Deep research and premium long-form answers are better handled by specialist models
Best suited to
Model routing and classification
Schema-constrained JSON output
Latency-sensitive background decisions
Short structured analysis
Aether Workspace Position
Available in Manywise on the Plus model list.
Manywise can use Mercury 2 only after deterministic filtering leaves an ambiguous choice, with a strict timeout and deterministic fallback.
Research sources
- Introducing Mercury 2Inception
- Mercury 2 on OpenRouterOpenRouter
