Mercury 3
Inception's diffusion-based language model. It generates more than 1,000 tokens per second and has a 1M-token context window.
Official site: inceptionlabs.ai ↗
Is it actually free? Free credits, then paid
New API accounts get 100 million free tokens. After that, you pay per token.
Pricing
$0.20 per 1M input tokens and $0.70 per 1M output tokens. At launch Inception cut both prices by 50%, to $0.10 input and $0.35 output, and has not given an end date. Cached input tokens cost 90% less than regular input.
You can use it on the Inception API Platform, OpenRouter, Baseten and Vercel. The API is OpenAI-compatible.
Best for
Apps where response time matters, such as voice and chat support, live tutoring, search agents and browser agents. Inception reports 1,068 tokens per second on NVIDIA GPUs.
Not great for
Teams that need independently checked quality. The claims that it beats GPT-6 Luna and Claude Haiku 5.5 come from Inception's own benchmarks.