MiniMax released M3.1 Flash, stating that agent-task costs have decreased 70%
BIBIBI
AT A GLANCE
MiniMax released M3.1 Flash, stating that its agent-task costs have decreased 70%, while online inference speed has increased by nearly 3 times.
Article
MiniMax released M3.1 Flash, reducing agent-task costs compared with the previous generation 70%
PANews October 20 According to reports,MiniMax the official blog,MiniMax officially released its new-generation model M3.1 Flash, reducing the cost of using agent tasks compared with the previous generation 70%, with online inference speed increasing by nearly 3 times and output speed exceeding per second 200 tokens；API is priced at per million input tokens 0.15 USD and output tokens 0.6 USD. The model achieved a score in the OSWorld V2 benchmark 61.7, ranking first among domestically developed models. According to the official statement, the anonymous model “Space Bunny,” which previously processed more than 200000000000000 tokens, was an early version of M3.1 Flash. The model is currently available through MiniMax API, M Plan, and MiniMax Code, and its model weights will be open-sourced later.MiniMax previously, in September launched...
PANews reported that MiniMax officially released its new-generation model M3.1 Flash。
02
According to the MiniMax official blog, the model's agent-task usage costs are lower than those of the previous generation 70%, and its online inference speed has increased by nearly 3 times.
03
The model's output speed exceeds per second 200 tokens。
04
API It is priced at per million input tokens 0.15 USD and per million output tokens 0.6 USD。
05
The model achieved a score in the OSWorld V2 benchmark 61.7, ranking first among domestically developed models.
06
According to the official statement, the anonymous model “Space Bunny” is an early version of M3.1 Flash, which previously processed more than 200000000000000 tokens。
07
M3.1 Flash It is now available through MiniMax API, M Plan, and MiniMax Code.
08
The model weights will be open-sourced later.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
M3.1 Flash is a newly released model. The main selling points cited in the report are lower costs for executing agent tasks, faster online inference, and pricing based on input and output tokens usage API. The model is currently available through multiple channels of MiniMax, but its weights have yet to be open-sourced.
Why it matters to readers
If the officially announced cost, speed, and benchmark results can be reproduced in actual use, developers may need to spend less to run agent tasks and obtain faster output.
Beginners should focus on API the separate billing for input and output, as well as actual task tokens usage; the performance and cost reductions cited in the report do not mean that all use cases can achieve the same results.
Risks and unknowns
The evidence is PANews' account of the MiniMax official blog; the original official blog post was not provided.
The methodology, compared previous-generation model, and applicable scenarios for the cost reduction 70%and nearly 3-fold speed increase are not specified.
OSWorld V2 score 61.7 and the evaluation scope and comparators for “first among domestically developed models” are not specified.
The specific timing and license for open-sourcing the model weights are not specified.
The end of the original text is truncated, making it impossible to confirm the content following “MiniMax previously, in September launched”.
Related Developments
Loading event timeline…
Related concepts
agent tasks
This term is not in the glossary yet. Browse related concepts in the glossary.
