Intel / video

MiniMax's Open Weight Model Strategy and Inference Stack

Source: youtube.com
Author: AI Engineer
Date: Published Jul 31, 2026

Context

An open model's serving performance can continue changing after its weights are released. In this discussion between MiniMax and Together AI, MiniMax describes making its model available for others to optimize, while Together AI explains the infrastructure work needed to serve it.

According to the panel, early architecture details let the serving team prepare GPU execution code before launch. Subsequent work targets processing, numerical precision and the cached information retained across an 's repeated requests. Long conversations and large codebases change what must be stored and moved, so improving the service involves the surrounding infrastructure as well as the model.

Terms in this piece · Glossary

attention — The mechanism that lets a model weigh which earlier words matter for the word it's currently processing — the core operation of a transformer.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
multimodal — A model that works with more than text — reading images, audio, or video, and sometimes generating them too.

Key quotes

“it aligns with our mission that we want to have intelligence with everyone”
Olive (MiniMax)

“it was trained multimodal from scratch. So, from step zero, we trained not only text data, we also trained image data.”
Olive (MiniMax)

“If someone tells me you can't do the thousand first thing, yeah, like I don't know, try harder.”
Dan (Together AI)

“It's like the type of thing that you should have done in your third year of undergrad or something like that, but most of us actually skipped that class, so now we're rediscovering it live in in industry.”
Dan (Together AI)

“we're we're seeing with models like M3 and GLM and Kimmy and and all those models that um the open-source frontier really can catch up”
Dan (Together AI)

Chapters

0:00Introducing the RL lead at MiniMax
1:17Why open source and open weights
3:45What builders are doing with the model
4:11A model that builds games
5:12Computer use and OS World
5:49Writing GPU kernels
6:25Parallel kernel bench
7:28A day zero inference stack
9:34Optimization across the stack
10:47Multimodality and training collapse
14:10Replicating a twelve hour run
17:22Where open models go next

Read the source www.youtube.com