Mercury 2.5 Preview is an early version of Inception's diffusion-based language model that generates and refines multiple tokens in parallel.
- Strengths
- Parallel token generation reduces latency compared to sequential autoregressive decoding, and the 260k context window handles substantially longer documents than earlier Inception models.
- Best for
- Applications where inference speed and long-context reasoning matter more than waiting for the stable release; preview versions carry the risk of API changes or discontinued support.
- Limitations
- As a preview release, this model may have rough edges, limited availability, or be superseded by Mercury 2.5 stable; users needing production stability should wait for the final release.
Input / 1M
$0.2
Output / 1M
$0.75
Cached input / 1M
$0.02
Context window
260K
Price history
Input (solid)Output (dashed)
Price change
- 30d
- in increased 400.0% out increased 400.0%
- 90d
- in increased 400.0% out increased 400.0%
- 1y
- in increased 400.0% out increased 400.0%
- Since launch
- in increased 400.0% out increased 400.0%
Snapshots
| Effective | Input | Output | Cached in | Note | Source |
|---|---|---|---|---|---|
| 8 Sep 2026 | $0.2 | $0.75 | $0.02 | Imported from OpenRouter | openrouter.ai |
| 1 Sep 2026 | $0.04 | $0.15 | $0.004 | Imported from OpenRouter | openrouter.ai |