# MiniMax launches M3.1 Flash, cutting agent costs by 70%

> MiniMax's M3.1 Flash model slashes agent task costs by 70% while boosting speed and security.
- Title: MiniMax launches M3.1 Flash, cutting agent costs by 70%
- Summary: MiniMax's M3.1 Flash model slashes agent task costs by 70% while boosting speed and security. MiniMax's M3.1 Flash model reduces agent task costs by 70%…
- Keywords: minimax, m3.1 flash, space bunny, agent tasks, cost reduction, ai models, chinese ai, finance, markets, launches, Flash, cutting
- Source: BigGo Finance — https://finance.biggo.com/news/4dc43d4f-0802-44a0-9a67-459a98cc892f
- Author: BigGo Finance
- Published: 2026-10-19T14:25:19+00:00
- Read time: 2 min
- Topics: minimax, m3.1 flash, space bunny, agent tasks, cost reduction, ai models, chinese ai, finance, markets
## MiniMax announces M3.1 Flash update
MiniMax announced an update to MiniMax M3.1 Flash on October 19, confirming 'Space Bunny' as an early version of the model.

> Space Bunny, an anonymous model that had processed more than 200 trillion tokens
## M3.1 Flash delivers generational gains
The new model shows generational improvements in agent tasks, coding, computer operations, and cybersecurity with online inference speed nearly three times faster.

> generational gains in agent tasks, coding, computer operation and cybersecurity
## Cost of agent tasks drops 70%
The overall cost of agent-type tasks decreased by 70% from the previous generation, with single enterprise workflow tasks costing 0.18 to 0.30 yuan.

> the overall cost of agent-type tasks down 70% from the previous generation
## M3.1 Flash benchmarks top Chinese models
On Automation-Bench v1.0.6, M3.1 Flash scored 62.2, above DeepSeek V4.1 Flash-1017's 60.5 and first among compared models.

> M3.1 Flash scored 62.2, above DeepSeek V4.1 Flash-1017's 60.5
## BankerToolBench results
M3.1 Flash scored 76.2 on BankerToolBench, tied with Claude Opus 5, showing strong multi-step planning and financial compliance capabilities.

> M3.1 Flash scored 76.2 on BankerToolBench
## OSWorld 2.0 performance
M3.1 Flash scored 61.7 on OSWorld 2.0, ranking first among Chinese models for computer operation and cross-application tasks.

> M3.1 Flash scored 61.7 on OSWorld 2.0
## Cybersecurity benchmark results
M3.1 Flash scored 82.2 on CyberGym, above Kimi K3's 80.0 but below GPT-5.6 Sol's 83.6 and GLM-5.3's 84.5.

> M3.1 Flash scored 82.2 on CyberGym
## Cost-effective enterprise tasks
For less than 0.30 yuan, M3.1 Flash completes enterprise workflow tasks, with average costs of 0.18 to 0.30 yuan per task.

> for less than 0.30 yuan, have the model attempt to complete an enterprise workflow task
## ALE benchmark results
M3.1 Flash achieved a 29.6% pass rate on ALE with an estimated total cost of 739 yuan, the lowest among models shown.

> M3.1 Flash achieved a 29.6% pass rate at an estimated total cost of about 739 yuan
## M3.1 Flash availability and pricing
M3.1 Flash is available via MiniMax API, M Plan, and MiniMax Code with pricing at 1 yuan per million input tokens and 4 yuan per million output tokens.

> M3.1 Flash at 1 yuan per million input tokens, 4 yuan per million output tokens
## Pro version in development
MiniMax plans to launch a Pro version with higher intelligence ceiling to pair with the Flash series for high efficiency and low cost.

> a Pro version aimed at a higher intelligence ceiling is in the works
## Key takeaway

MiniMax's M3.1 Flash model reduces agent task costs by 70% compared to previous versions, with online inference speed nearly three times faster.