Mercury 3 by Inception: diffusion AI at 1,000 tokens per second
Mercury 3, by Inception, writes about 1,000 tokens per second, reads 1M of context and costs $0.10 per million input tokens at launch.
Mercury 3 is the Inception model that writes 1,000 tokens per second and reads a million tokens of context. Inception introduced it on October 19, 2026. It is a diffusion model: it doesn't generate word by word like traditional models. It is compatible with the OpenAI API and comes with 100 million free tokens on Inception's platform to try it.
What Mercury 3 is
If your customer is waiting for the answer, every second costs a sale. Mercury 3 targets exactly that: speed. It generates by diffusion, not word by word, and reaches about 1,068 tokens per second according to Inception.
What it does
- It writes about a thousand tokens per second.
- It reads a million tokens of context.
- It plugs in like the OpenAI API, so it works with code that already uses that format.
- It claims to match Haiku 5.5, four to eight times faster.
Price and fine print
The launch price is $0.10 per million input tokens and $0.35 per million output tokens. The list price is $0.20 and $0.70. That makes the launch discount 50%, and the announcement doesn't say when it ends. Also, the speed and the comparison with other models are the company's own figures, not independent tests.
How to apply it in your business?
- Identify the tasks that can't wait: customer replies, live assistants, any flow where latency costs sales.
- Try it for free with the 100 million tokens on Inception's platform.
- Reuse your code: since it's compatible with the OpenAI API, a test migration is simple.
- Measure speed and quality yourself before trusting the comparison with Haiku 5.5.
- Don't plan around the discount: it has no known end date, so budget with the list price.
If you compare prices per million tokens, read the full breakdown of DeepSeek V4.1 Flash.
Frequently asked questions
What is Mercury 3?
A diffusion model from Inception, introduced on October 19, 2026, that writes about 1,000 tokens per second and reads a million tokens of context.
How much does it cost?
At launch, $0.10 per million input tokens and $0.35 output; the list price is $0.20 and $0.70.
Is the discount permanent?
The announcement doesn't say when the launch discount ends.
Is it compatible with the OpenAI API?
Yes, it plugs in like the OpenAI API.
Can I try it for free?
Yes, with 100 million free tokens on Inception's platform.
Conclusion
Mercury 3 bets on speed: diffusion, a thousand tokens per second, a million of context and a launch price of $0.10 per million input tokens. The speed figures and the Haiku 5.5 comparison are the company's own, so test it on your tasks before adopting it. Source: Inception, "Introducing Mercury 3" (October 19, 2026).