Mercury 2.5 AI Evaluation | LayerLens - Benchmark Results & Performance

Model

Mercury 2.5 by Inception

Text->text

Free credits included. No credit card required.

Mercury 2.5 is the most capable diffusion large language model (dLLM) from Inception and the fastest reasoning LLM in production. Unlike autoregressive models that generate tokens sequentially, Mercury 2.5 produces and refines multiple tokens in parallel, achieving over 1,100 tokens per second on widely-available NVIDIA GPUs. It features a 260K token context window, tunable reasoning, parallel tool calls, and schema-aligned JSON output. Mercury 2.5 delivers intelligence comparable to cost-optimized frontier models while maintaining exceptional throughput and low cost.

Evaluations

0

No evaluations run yet

Latency

—

Context length

260k

tokens

Cost

$0.04 · $0.15

input · output per 1M tokens

Key takeaways

Mercury 2.5 is designed as a diffusion large language model (dLLM) for high-speed text generation and reasoning.
It aims to solve the problem of slow token generation in traditional autoregressive models by producing and refining tokens in parallel, achieving over 1,100 tokens per second.
The model is a Transformer-based architecture with a 260K token context window, tunable reasoning, parallel tool calls, and schema-aligned JSON output.
A key capability is its exceptional throughput and low cost, delivering intelligence comparable to cost-optimized frontier models while maintaining high speed.
No specific evaluation results for best and worst cases were provided in the model description.

Technical specifications

Context length: 260k tokens
Max output: —
Architecture: Transformer
Modalities: text->text
Parameters: —
Open weights: No
Licence: Proprietary
Regions: USA
Vendor: Inception
Provider model key: inception/mercury-2.5