Inception's Mercury 2.5, a diffusion model at $0.75 per million output tokens
Inception released Mercury 2.5 on 9 September 2026[1]. It calls it the most capable diffusion language model on the market and, to its knowledge, the largest one ever trained[1].
Most language models write left to right, one token at a time. A diffusion model produces many tokens together and refines them over several passes.
The context window is 260K tokens[1]. Inception puts its speed at 1,107 tokens per second on widely available NVIDIA GPUs[1].
The standard price is $0.20 per million input tokens and $0.75 per million output tokens[1]. At launch it is 80 percent off, at $0.04 and $0.15, and the announcement gives no end date for the discount[1].
Inception describes its quality as comparable to cost-optimized frontier models such as GPT-5.6 Luna (Low), Gemini 3.5 Flash-Lite and Claude Haiku 4.5[1]. It does not say what the comparison is based on[1].
For reference, gpt-5.6-luna is $0.20 input and $1.20 output, and Claude Haiku 4.5 is $1 and $5[2,3]. At the standard price Mercury 2.5 matches luna on input and costs nearly 40 percent less on output.
It supports tunable reasoning, parallel tool calls and schema-aligned JSON[1]. It is available through Inception's own API, Baseten and OpenRouter[1].