Home
MiniMax Apologizes for Poor Communication on New Billing Model, Rolls Out Compensation and Refunds

MiniMax, an AI tech company, recently shifted its billing structure from a per-task model (Coding Plan) to a token-based system (Token Plan) with the launch of its new flagship model, MiniMax-M3. Because users were not informed in advance and token consumption far exceeded expectations, many heavy users quickly burned through their monthly quotas, sparking widespread dissatisfaction and a wave of complaints from the developer community.
Under mounting public pressure, MiniMax issued an apology on the evening of June 2, acknowledging that the change was communicated poorly, the transition plan was insufficiently thought through, and the treatment of existing users' weekly limits was mishandled. The company explained that the new M3 model is larger, supports native multimodal capabilities, and offers a context length of up to one million tokens, requiring greater computational resources. The move to token-based billing, they added, was intended to allow users to allocate their subscription quotas freely across different modalities.
To make amends and reward subscribers, MiniMax announced the following compensation and benefit adjustments:
Old User Benefits Preserved: Users who purchased before March 22, 2026, and originally enjoyed unlimited weekly usage will retain that privilege for both M2.7 and M3 models after the upgrade.Permanent Weekly Limit Increase: Users who purchased the Token Plan before 10:00 AM on June 5 will receive a permanent 50% increase in their weekly M3 model quota for the duration of their subscription.Temporary Quota Doubling: The company reset all user quotas. From June 1 through June 7, all subscribers' weekly usage limit of "5 hours" will be temporarily doubled.Extended Points and Refund Channel: The validity of previously issued compensation points has been automatically extended from one month to one year. Additionally, an online self-service refund portal is expected to launch on June 3 (Tuesday), enabling users to submit refund requests directly.This incident is not unique. Earlier, Moonshot (Kimi) encountered similar user backlash over billing model changes. Industry experts note that as large models move toward AI agent scenarios, token usage per task has grown exponentially, making token-based billing an inevitable strategy for companies to manage computing costs. Yet, during this commercial transition, safeguarding users' right to information and building a comprehensive consumer trust framework remain critical challenges for the industry.
Related article
Pichai Confirms Google’s Gemini 4 Is Now in Training, With Reveal Expected by Year-End
Sundar Pichai, Google’s CEO, confirmed during the company’s 2026 second-quarter earnings call that training for the next-generation Gemini4 model has officially begun, with Google committing “a significant amount of resources” to its development. Bas
Xiaomi 18 Pro Series Launches With Camera-Level Optical Bokeh Powered by End-to-End Large Model
Xiaomi has confirmed that the Xiaomi 18 Pro series will launch on September 23 at 7 PM, featuring two variants: the 18 Pro and 18 Pro Max. The standout innovation is the new "Legendary Instant" imaging system, which leverages an end-to-end large mode
Apple Music to Label AI-Generated Content as Music Platforms Tackle AI Spam
Starting in late 2026, Apple Music will mandate AI transparency labels for content providers, transitioning from a voluntary disclosure system to strict enforcement. Since introducing these labels in March 2026, Apple has focused on providing listene
Related Special Topic Recommendations
Comments (0)
0/500

MiniMax, an AI tech company, recently shifted its billing structure from a per-task model (Coding Plan) to a token-based system (Token Plan) with the launch of its new flagship model, MiniMax-M3. Because users were not informed in advance and token consumption far exceeded expectations, many heavy users quickly burned through their monthly quotas, sparking widespread dissatisfaction and a wave of complaints from the developer community.
Under mounting public pressure, MiniMax issued an apology on the evening of June 2, acknowledging that the change was communicated poorly, the transition plan was insufficiently thought through, and the treatment of existing users' weekly limits was mishandled. The company explained that the new M3 model is larger, supports native multimodal capabilities, and offers a context length of up to one million tokens, requiring greater computational resources. The move to token-based billing, they added, was intended to allow users to allocate their subscription quotas freely across different modalities.
To make amends and reward subscribers, MiniMax announced the following compensation and benefit adjustments:
Old User Benefits Preserved: Users who purchased before March 22, 2026, and originally enjoyed unlimited weekly usage will retain that privilege for both M2.7 and M3 models after the upgrade.Permanent Weekly Limit Increase: Users who purchased the Token Plan before 10:00 AM on June 5 will receive a permanent 50% increase in their weekly M3 model quota for the duration of their subscription.Temporary Quota Doubling: The company reset all user quotas. From June 1 through June 7, all subscribers' weekly usage limit of "5 hours" will be temporarily doubled.Extended Points and Refund Channel: The validity of previously issued compensation points has been automatically extended from one month to one year. Additionally, an online self-service refund portal is expected to launch on June 3 (Tuesday), enabling users to submit refund requests directly.This incident is not unique. Earlier, Moonshot (Kimi) encountered similar user backlash over billing model changes. Industry experts note that as large models move toward AI agent scenarios, token usage per task has grown exponentially, making token-based billing an inevitable strategy for companies to manage computing costs. Yet, during this commercial transition, safeguarding users' right to information and building a comprehensive consumer trust framework remain critical challenges for the industry.
Pichai Confirms Google’s Gemini 4 Is Now in Training, With Reveal Expected by Year-End
Sundar Pichai, Google’s CEO, confirmed during the company’s 2026 second-quarter earnings call that training for the next-generation Gemini4 model has officially begun, with Google committing “a significant amount of resources” to its development. Bas
Xiaomi 18 Pro Series Launches With Camera-Level Optical Bokeh Powered by End-to-End Large Model
Xiaomi has confirmed that the Xiaomi 18 Pro series will launch on September 23 at 7 PM, featuring two variants: the 18 Pro and 18 Pro Max. The standout innovation is the new "Legendary Instant" imaging system, which leverages an end-to-end large mode
Apple Music to Label AI-Generated Content as Music Platforms Tackle AI Spam
Starting in late 2026, Apple Music will mandate AI transparency labels for content providers, transitioning from a voluntary disclosure system to strict enforcement. Since introducing these labels in March 2026, Apple has focused on providing listene











