The efficiency argument: most agent calls do not need frontier intelligence
- Source
- x.com
- Date
"Beyond autoregressive: why diffusion is the future of language models." @StefanoErmon's keynote at @StartupGrind last week at the Fox Theatre. Your AI product doesn't make one model call per session. It makes thousands. Most run quietly under the hood. The work that keeps the agent moving. Using a frontier model for all of it means paying frontier pricing for work that doesn't need frontier intelligence. Mercury 2 is built for that work: >1,000 tokens/sec on standard GPUs, comparable quality to frontier speed-optimized models for a fraction of the cost. The question isn't which model is smartest. It's which model is most efficient, on the highest-volume tasks. @adityagrover_ @volokuleshov
Context
Inception's efficiency argument is that model selection should account for the recurring work inside an AI product. A system that calls a model repeatedly pays the cost and waits for the response each time. The company argues that many supporting tasks need less capability than the most demanding task in the same application, and positions Mercury 2 for that work. The resulting design choice is to evaluate whether a faster, cheaper model handles a frequent task well enough to justify assigning that task to it. Company argument
postInception ships Mercury Decide update with ZDR and 66k context
postInception launches Mercury Voice, a diffusion LLM for low-latency voice agents
postToday, we’re introducing Mercury 2.5, the most capable diffusion LLM on the…
postInception ships Mercury 2.5 Preview at roughly 1,100 tokens per second
Checking sign-in…
Loading comments…