Fast and very cheap during its 80% launch discount, it suits developers after Haiku-class answers at higher speed.

Kenning's take

What it does

Mercury 2.5 is Inception's most capable diffusion language model, which the company says is the largest ever trained. Inception reports a 40% rise in intelligence over Mercury 2, putting it near cost-focused models such as Claude Haiku 4.5, while running at about 1,107 tokens per second. It supports tunable reasoning, parallel tool calls and structured JSON, and is available in the Inception API as mercury-2.5.

• 1,107 tokens per second on NVIDIA GPUs (Inception figure)
• 260K token context
• Tunable reasoning effort
• Parallel tool calls and schema-aligned JSON
• Used for search, voice and coding subagents

Who it suits. Developers who need a fast, low-cost model for search pipelines, voice agents and coding subagents.

Key facts

Category

AI models

Part of

Mercury 2 (Inception)

Made by

Inception

Released

8 September 2026

How to get it

through the Inception API as mercury-2.5

Works on

API

API price

$0.20 per million input tokens, $0.75 per million output (launch offer $0.04 and $0.15)

FAQ

Facts from the maker's own pages and the sources listed. Checked 2026-10-11.