MiniMax Token Plan

MiniMax Token Plan is the current subscription plan for MiniMax coding and agent workflows. It replaces the older prompt-count coding-plan framing with a shared credit-based quota for all supported MiniMax API Platform resources.

The headline model is now MiniMax-M3, released on June 1, 2026. MiniMax positions M3 as its latest M-series model for coding, tool use, agentic reasoning, multimodal chat input, and long-context work with a 1,000,000-token context window.

Token Plan usage works through a Subscription Key. The same included quota can be used across supported text, image, speech, music, video, and MCP resources, with the console usage bar as the source of truth for remaining capacity.

• Plus: 20 USD/month, about 1.633B monthly M3 tokens
• Max: 50 USD/month, about 5.053B monthly M3 tokens
• Ultra: 120 USD/month, about 9.796B monthly M3 tokens

MiniMax also sells Credits packages that can cover eligible overflow after the included Token Plan quota is used:

• 5 USD: 6,000 credits
• 25 USD: 32,000 credits
• 100 USD: 140,000 credits

Credits are listed at 1,000 credits = 1 USD, are valid for 365 days from purchase, and share the same resource coverage as Token Plan.

• Public tiers are Plus, Max, and Ultra. The earlier Starter and High-Speed prompt-count tiers are no longer the current public Token Plan table.
• Included Token Plan quota is controlled by both a 5-hour rolling window and a weekly window.
• Usage is deducted by actual resource consumption. Simple chat or editing tasks should consume less than long-context, multimodal, or complex multi-agent workflows.
• Typical concurrent agent capacity is listed as about 3-4 agents on Plus, 4-5 agents on Max, and 6-7 agents on Ultra.
• Usage from all supported tools and resources shares the same included quota. Purchased Credits can cover eligible overflow.
• Unused included Token Plan quota does not carry over to the next billing cycle.
• MiniMax recommends pay-as-you-go rather than Token Plan for production workloads.

M3 makes MiniMax Token Plan more relevant for coding-agent users than the older M2.7 prompt plan, especially if you need long context or multimodal inputs. The tradeoff is that the plan is less predictable than a fixed prompt counter: usage now depends on actual token/resource consumption and is governed by both 5-hour and weekly windows. Treat the console usage bar as authoritative, and assume large codebase scans or autonomous agent loops can burn quota quickly.

• June 1, 2026: MiniMax-M3 released and Token Plan updated around shared credit-based usage for Plus, Max, and Ultra tiers.
• March 18, 2026: MiniMax-M2.7 released.
• 2026 Q1: Earlier MiniMax coding-plan pricing used prompt/request counts and separate High-Speed tiers.