# Why AI Agents Need Million-Token Context — Thomas Wolf & Olive Song, MiniMax

Source: [AI Engineer](https://www.youtube.com/watch?v=5Cxe5dv2Xlw)  
Feed7 permalink: https://feed7.dev/p/why-ai-agents-need-million-token-context-thomas-wolf-olive-song-minimax-02jmtmy  
Published: 2026-09-04T13:00:26.000Z  
Trust: Source Linked (source_linked)

## Why Included

MiniMax argues that long-running, multimodal agents need enough context for repeated tool responses, while sparse attention makes that context more tractable.

## Source Summary

MiniMax M3 has roughly **400B total parameters**, activates **20B**, and supports text, images, and video. Its **1M-token context** uses MiniMax Sparse Attention: an index branch selects relevant blocks, then a sparse branch performs attention over them.

## Practical Implication

For agent builders, the practical claim is that multi-round tool responses and long, unstructured inputs can exceed short contexts. M3’s native multimodal training also points toward agents that inspect reports, presentations, tutorials, images, and video before acting with tools.

## Agent-Ready Context

MiniMax M3 has roughly **400B total parameters**, activates **20B**, and supports text, images, and video. Its **1M-token context** uses MiniMax Sparse Attention: an index branch selects relevant blocks, then a sparse branch performs attention over them.

For agent builders, the practical claim is that multi-round tool responses and long, unstructured inputs can exceed short contexts. M3’s native multimodal training also points toward agents that inspect reports, presentations, tutorials, images, and video before acting with tools.

A large context window does not establish useful recall across its full length, and the discussion supplies no retrieval or agent-task measurements. Trillion-token attention is presented only as a research direction requiring architecture and hardware work.

## Connected Context

Feed7 judgment across 691 accumulated Signals:

This adds an architectural explanation for how a 1M-token multimodal model can make large contexts computationally practical and clarifies why tool-heavy agents may need them. Against the candidate model releases, it confirms that million-token capacity is becoming common but does not resolve model selection: neither sparse attention nor native multimodality demonstrates reliable recall or better agent performance without workload-level evaluation.

- [Hy4 Preview now available on AI Gateway](https://feed7.dev/p/hy4-preview-now-available-on-ai-gateway-0knvicj) — Both add open MoE models with 1M-token contexts to coding-agent choices; M3 contributes a specific sparse-attention design, while neither supplies sustained-context performance evidence.
- [GLM 5.3 now available on AI Gateway](https://feed7.dev/p/glm-5-3-now-available-on-ai-gateway-0s7o9zv) — GLM 5.3 reinforces that million-token capacity alone is not differentiating; its long-horizon claims and M3’s architectural claims both still require matched agent-task evaluation.
- [Gemini 3.8 Flash now available on AI Gateway](https://feed7.dev/p/gemini-3-8-flash-now-available-on-ai-gateway-0cun77q) — Gemini’s million-token multimodal route confirms the same emerging capacity envelope, making recall quality, tool behavior, latency, and token use more useful selection criteria than context size alone.

## Context Map

- Layer: model
- Domains: coding, image
- Topics: open-models, reasoning, coding-agents

## Uncertainty

- A large context window does not establish useful recall across its full length, and the discussion supplies no retrieval or agent-task measurements. Trillion-token attention is presented only as a research direction requiring architecture and hardware work.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.