---
format: "aidr-story-markdown/v1"
id: "a9e40bcc771174092a201e91e069e6d804439d5a1227359ef5064a38466e9d47"
canonical_url: "https://aidr.today/a9e40bcc?lang=en"
title: "Kimi K3 (2.8T) at 1 token/s on a MacBook Pro, streamed from four SSDs"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-09-08T20:07:55.000Z"
category: "Research"
topics: ["kimi","llm","inference","apple"]
source_urls: ["https://github.com/argonautlabsai/deltafin","https://news.ycombinator.com/item?id=49616257"]
summary: "ARGODRIVE Deltafin: Kimi K3 (2.8T MoE) streamed from SSDs on Apple Silicon — fork of gavamedia/deltafin with the ARGODRIVE storage work and benchmark package - argonautlabsai/deltafin"
---

# Kimi K3 \(2\.8T\) at 1 token/s on a MacBook Pro, streamed from four SSDs

> [Open the canonical story](<https://aidr.today/a9e40bcc?lang=en>)

**Published:** 2026-09-08T20:07:55.000Z
**Category:** Research
**Topics:** kimi, llm, inference, apple

## Summary

ARGODRIVE Deltafin: Kimi K3 \(2\.8T MoE\) streamed from SSDs on Apple Silicon — fork of gavamedia/deltafin with the ARGODRIVE storage work and benchmark package \- argonautlabsai/deltafin

## Sources

- [Story source](<https://github.com/argonautlabsai/deltafin>)
- [Discussion](<https://news.ycombinator.com/item?id=49616257>)