MiniMax
The lab behind Hailuo, now shipping agent-grade Flash models at token prices a fraction of the usual — M3.1 Flash launched October 2026.
- Web
- Why it's listed
- The same lab built Hailuo, one of the strongest Chinese video generators we list.
- Consider before using
- The consumer chat UI is Chinese-first, which can block English-only users.
From the vendor's site: MiniMax is the Chinese AI lab behind the Hailuo video generator and the Talkie companion app. Its M-series flagship models target agentic work — coding, computer use, cybersecurity — and the October 2026 M3.1 Flash delivers roughly three times faster online inference and seventy percent lower cost than its predecessor, at $0.15 input and $0.60 output per million tokens.
What it is
MiniMax is a major Chinese AI lab. Most readers will know it through Hailuo (its video generator, covered in our video guide) or Talkie, but the company's core line is the M-series of text models built for agents: long task chains, tool calls, coding and computer use. In October 2026 it added the Flash tier to make that capability affordable.
What's new: M3.1 Flash (October 2026)
MiniMax launched M3.1 Flash on October 19, 2026. It previously ran anonymously as "Space Bunny" and processed over 200 trillion tokens before the public name. Per the official launch notes, it makes online inference nearly three times faster and cuts usage cost by seventy percent versus its predecessor.
Pricing per 1M tokens (official launch notes):
| Item | Price |
|---|---|
| Input | $0.15 |
| Output | $0.60 |
| Cached input | $0.015 |
Official benchmark results from the launch post:
- 73.7% on DeepSWE v1.1 (real-world software engineering tasks)
- 61.7 on OSWorld V2 and 29.6 on Agents' Last Exam — first among Chinese models on both at launch
- 76.2 on BankerToolBench for multi-step, document-consistent financial workflows
- 82% on CyberGym for vulnerability discovery and validation
- Output speed: 200+ tokens per second on everyday use, with a median of 170+
A higher-end Pro tier was announced as coming later; model weights will also be released later.
How to use it
There are three doors: the token-priced API at platform.minimax.io, the subscription "M Plan", and the MiniMax Code desktop agent (download at agent.minimax.io). There is also a browser-based MiniMax Agent app at chat.minimax.io.
Verdict
Pick MiniMax when the job is agents and automation on a budget: M3.1 Flash brings leader-level benchmark performance to a $0.15/$0.60 per million token price set. For everyday chat, DeepSeek remains the stronger all-rounder, and for video the lab's Hailuo is the better door.
Last checked: October 20, 2026. Pricing and features change fast — confirm current numbers on the official site.