Inception's Mercury 2.5 Diffusion LLM Lists at $0.20/$0.75 per 1M Tokens, 80% Off at Launch
- What happened
- Inception released Mercury 2.5, a diffusion LLM, at $0.20/$0.75 per 1M tokens, 80% off at launch with no end date.
- Why it matters
- Inception claims 1,107 tokens per second and quality comparable to Gemini 3.5 Flash-Lite, at a lower list price on both lines; no benchmark scores back it.
- What to do
- Test it on a latency-bound workload while the discount lasts, and cost it at the list rate.
Our verdict on Mercury 2.5 is Conditional: test it if your workload is high-volume and latency-bound, where raw speed is the reason to pick it, and the launch discount makes that test nearly free. Inception has given no end date for the discount, so price anything long-running at the list rate.
What happened
Inception published "Introducing Mercury 2.5", signed by CEO Stefano Ermon. Mercury generates text by diffusion rather than one token at a time, and Inception calls 2.5 "the most capable diffusion LLM on the market".
- Price. "$0.20 per million input and $0.75 per million output". "At launch, Mercury 2.5 is 80% off", which brings it to $0.04 input and $0.15 output.
- Context. 260K tokens.
- Speed. "1,107 tokens per second on widely-available NVIDIA GPUs".
- Where. "Mercury models are available through our Inception API, Baseten, and OpenRouter."
Why it matters
Inception says Mercury 2.5 is a "40% increase in intelligence from Mercury 2" and places it as comparable to "cost-optimized frontier models like GPT-5.6 Luna (Low), Gemini 3.5 Flash-Lite, and Claude Haiku 4.5". The post names no benchmark scores for either claim; it says the team used customer feedback and production failure cases to sharpen its evals. We have not run it.
For scale, Gemini 3.5 Flash-Lite, one of the models Inception names, lists at $0.30 input and $2.50 output in our directory. If Mercury 2.5 really matches it on quality, its list price is lower on both lines, and its claimed speed is roughly three times the 350 tokens per second we list for Flash-Lite.
What changes for you
- Latency-bound, high-volume jobs: run them through Mercury 2.5 on the Inception API or OpenRouter while the 80% discount lasts.
- Long-running budgets: cost them at $0.20/$0.75, not the launch rate.
- Quality-sensitive work: treat the "comparable to Flash-Lite and Haiku 4.5" claim as unproven until you test it on your own evals.
FAQ
How long does the 80% launch discount last? Inception gives no end date, so assume it can end at any time.
Where can I use Mercury 2.5? Through the Inception API, Baseten and OpenRouter, according to Inception.
What to do
- 1 Run a latency-bound workload through Mercury 2.5 on the Inception API or OpenRouter while the 80% launch discount lasts.
- 2 Price any long-running use at the $0.20/$0.75 list rate: Inception gives no end date for the discount.
- 3 Check the quality claim against your own evals before moving work off Gemini 3.5 Flash-Lite or Claude Haiku 4.5.
Affected tools & models
Never need to catch up again
The weekly delta: only verdict changes and act-now items. No digest filler.