Inception introduced Mercury 3, a language model that generates text through diffusion rather than word by word. The company reports 1,068 tokens per second on Nvidia GPUs and a one-million-token context window. According to its own tests, it beats speed-focused models such as GPT-6 Luna and Claude Haiku 5.5. Pricing is $0.20 per million input tokens, with 50% off at launch.

What it means: Most language models write text one piece at a time, which caps their speed. Diffusion models generate and refine many pieces at once, the way image models do. Inception argues this approach now scales as well as the traditional one and that quality is close to the fast models from the big labs. The figures come from the company itself, so independent measurements are needed. Speed does have practical value, though. For agents that browse websites or analyse files, every step waits on the model, and small delays add up. The same applies to voice assistants and customer service, where a few seconds of waiting spoils the experience. For Greek businesses building chatbots or automations, cheap and fast alternatives lower the cost per conversation. Such models are worth testing in Greek as well, because performance in smaller languages often differs from English. Competition on speed will also put pressure on the large providers.

Read the original on Inception

Related service: Custom software

Related stories