Need a China coding plan (GLM / MiniMax / Kimi / Volcengine / Xiaomi)? Email me at [email protected]

Back to comparison

MiniMax · Token Plan

MiniMax Token
Plan Guide

M3 flagship with native multimodal input and 1M context. Three international tiers on sale (Plus $20 / Max $50 / Ultra $120 per month — the legacy Starter and Max HS tiers were retired). Annual billing gives 2 months free, and three credit packs are available if the monthly quota runs out.

$20Entry / month
3Tiers
12.5BUltra tokens/mo
1MContext

Models

M3, M2.7 and H3 share the same monthly token quota. Music models were discontinued on Aug 20.

MiniMax M3

Flagship

Flagship model with native multimodal input (image & video), 1M context, ideal for large codebases.

M2.7 / M2.7-highspeed

High speed

Shares the monthly quota with M3. Standard ~50 TPS, high-speed ~100+ TPS. Reference limits per 5 hours by tier (Plus / Max / Ultra): 1,500 / 4,500 / 15,000 calls, or 750 / 2,250 / 7,500 on the high-speed build.

MiniMax H3

Open source

Reasoning and agent-orchestration model sharing quota with M3; open-sourced on Aug 3.

Image / Speech family

M2.5 and Speech 2.8 share the same monthly quota. Note: all Music models were discontinued on Aug 20, 2026 (the open-source MiniMax-Music3 remains on HuggingFace).

Plans

Monthly token usage model. Annual billing = 2 months free. Legacy Starter ($10) and Max HS ($80) tiers are retired and only kept for existing subscribers; top-up credit packs cover usage beyond the monthly quota.

Plus

$20/mo

$220/year

  • ~1.7B tokens/mo (M3)
  • 3-4 concurrent Agents
  • Multimodal models share quota
Best for

Individual developers and daily workflows

Subscribe to Plus →
Recommended

Max

Top pick
$50/mo

$550/year

  • ~5.1B tokens/mo
  • 4-5 concurrent Agents
  • Hailuo 2.3 video: 3/day
Best for

Daily professional developers

Subscribe to Max →

Ultra

$120/mo

$1,320/year

  • ~12.5B tokens/mo
  • 6-7 concurrent Agents
  • Hailuo 2.3 video: 5/day
Best for

Multi-project and heavy-usage scenarios

Subscribe to Ultra →

支持的编程工具

Claude CodeCursorClineOpenCodeMiniMax Code

Highlights

Native multimodal

M3 understands images and videos in addition to text.

MiniMax Code app

Official agent app (desktop + web): Token Plan subscribers can use it without an API key.

Top-up credit packs

Starter $5 (5,000 credits), Advanced $25 (25,000 credits) and Premium $100 (100,000 credits): valid for one year, stackable, and they automatically top up usage once the plan quota runs out.

Dual quota windows

The monthly quota is capped by both a fixed 5-hour window and a weekly window; unused quota does not roll over.

Peak-hour throttling

Dynamic throttling usually appears on weekdays between 15:00 and 17:30.

Frequently Asked Questions

How does the token quota work?

Plans use a monthly token-based quota: Plus ~1.7B, Max ~5.1B, Ultra ~12.5B tokens per month for M3 usage. Actual consumption depends on context length; image and speech models share the same quota.

What happened to the Starter and Max HS tiers?

Both were retired. Starter ($10) and Plus-hs ($40) remain only as legacy tiers for existing subscribers; Max HS ($80) users were migrated to the new Max tier ($50). The current lineup is Plus $20 / Max $50 / Ultra $120.

Can I still generate music?

No. All Music models stopped serving on Aug 20, 2026 in Token Plan and pay-as-you-go. The open-source MiniMax-Music3 weights are available on HuggingFace for self-hosting.

Can I top up when the plan quota runs out?

Yes. Three credit packs are available: Starter $5 (5,000 credits), Advanced $25 (25,000 credits) and Premium $100 (100,000 credits). They are valid for one year from purchase, can be stacked, and can automatically top up the plan quota once it is exhausted; the monthly quota itself is capped by a 5-hour window and a weekly window, and unused quota does not roll over.

Is annual billing worth it?

Annual plans cost $220 (Plus) to $1,320 (Ultra), effectively 2 months free compared to monthly billing.