models/inception/mercury-2
I
inception·active

Inception: Mercury 2

inception's efficiency model. Context window: 128K tokens.

Overall score
4.08
/5.00 · ranked #85
Input
$0.250
per 1M tokens
Output
$0.750
per 1M tokens
Context
128K
tokens
Blended
$0.625
3:1 out:in ratio

Price drops, new benchmarks, model updates. Stay current on Inception: Mercury 2.

One email per change. Unsubscribe anytime.

modelpicker.aipowered by live benchmark data

Scores by test

Methodology →
Structured Output
5.0
Strategic Analysis
5.0
Constrained Rewriting
3.0
Creative Problem Solving
4.0
Tool Calling
3.0
Faithfulness
5.0
Classification
4.0
Long Context
5.0
Safety Calibration
1.0
Persona Consistency
4.0
Agentic Planning
4.0
Multilingual
5.0
Tabular Data
5.0

What you need to know

Inception: Mercury 2 is a high-precision model optimized for structural integrity and analytical depth. It achieves perfect scores in structured output, faithfulness, and strategic analysis, making it highly reliable for tasks requiring strict adherence to formats and factual grounding. Its capability extends to complex data handling, performing at a top tier for tabular data, multilingual processing, and long-context retrieval across its 128K window.

Pricing is aggressive for a model with these capabilities, with a blended cost of $0.625/MTok. This puts it in a budget-friendly tier relative to its performance in high-logic domains like agentic planning and strategic analysis. However, the model ranks #73 of 105 overall, suggesting that while it excels in specific technical niches, it lacks the general-purpose versatility of higher-ranked models.

The model has critical failures in safety calibration, scoring 1/5, which indicates a lack of robust guardrails. It also shows mediocre performance in tool calling and constrained rewriting, meaning it may struggle with external API integrations or strict stylistic modifications.

Use this model if you need a low-cost solution for structured data extraction, strategic analysis, or multilingual processing where safety filtering is handled by your own middleware. Skip this model if your application requires high safety standards, reliable tool calling, or precise rewriting constraints.

Strengths — Top 3

Structured Output5.0/5.0
Strategic Analysis5.0/5.0
Faithfulness5.0/5.0

Relative weaknesses — Bottom 3

Safety Calibration1.0/5.0
Constrained Rewriting3.0/5.0
Tool Calling3.0/5.0

Similar models

QQwen: Qwen3.5-35B-A3B$0.7854.31BByteDance Seed: Seed 1.6$1.564.31OOpenAI: gpt-oss-120b$0.1374.08SStepFun: Step 3.5 Flash$0.2504.23