Inception: Mercury 2.5 Preview
inception's mid-tier model. Context window: 260K tokens.
Scores by test
Methodology →What you need to know
Inception: Mercury 2.5 Preview is optimized for high-reliability agentic workflows, excelling in structured output, tool calling, and agentic planning. With a 260K context window and perfect internal scores for long-context handling, tabular data, and multilingual support, it is built for complex data extraction and automation tasks that require strict adherence to formatting.
The model is priced aggressively for its performance tier, with a blended cost of $0.122/MTok. This makes it a high-value option for developers who need frontier-level capabilities in creative problem solving and faithfulness without the premium cost associated with top-tier proprietary models.
The primary trade-off is in safety calibration, where it scores a 2/5. In practice, this indicates a tendency to over-refuse benign prompts, which may introduce friction in user-facing applications. While it performs well in strategic analysis and rewriting, these are its weakest relative areas compared to its perfect scores in technical execution.
Use this model if you are building autonomous agents, complex data pipelines, or multilingual tools that require precise structured output. Skip this model if your application requires a highly permissive safety profile or if you cannot tolerate false-positive refusals in your user interactions.
Strengths — Top 3
Relative weaknesses — Bottom 3
Similar models