← Models

MiniMax M2.7

MiniMax · MiniMax M

MiniMax's M2.7 agentic productivity model for coding, office workflows, tool use, and low-cost long-context execution.

Part of MiniMax M family · Other versions: MiniMax M2.5 , MiniMax-M2.5
Type
language
Context
205K tokens
Max Output
205K tokens
Status
current
Input
$0.3/1M tok
Output
$1.2/1M tok
API Access
Yes
License
proprietary
chinese coding agentic tool-use long-context cost-efficient
Released March 2026 · Updated May 16, 2026

Overview

Freshness note: Model capabilities, limits, and pricing can change quickly. This profile is a point-in-time snapshot last verified on May 16, 2026.

MiniMax M2.7 is the current top M-series productivity model in MiniMax’s public docs. It follows M2.5 and keeps the same basic value proposition: inexpensive hosted intelligence for coding, tool use, office-style deliverables, and long-running agent workflows.

The March 18 launch frames M2.7 around recursive self-improvement and agent harness work. The more practical Signal Lens takeaway is simpler: MiniMax now documents M2.7 in release notes, model overview, pay-as-you-go pricing, token plans, and rate limits, so it is no longer just a watch item.

Capabilities

MiniMax positions M2.7 for complex agentic productivity tasks. The official model overview highlights top real-world engineering, professional office delivery, and character-rich interaction, while the launch post emphasizes agent harnesses, complex skills, dynamic tool search, and self-improving workflows.

That makes it most relevant for coding agents, office-document automation, research assistants, and low-cost long-context loops where a premium western model would be overkill or too expensive to run continuously.

Technical Details

Public anchors at this snapshot:

  • Standard model ID: MiniMax-M2.7.
  • High-speed sibling: MiniMax-M2.7-highspeed, documented as the same capability with faster inference.
  • API overview lists a 204,800 total-token maximum for M2.7.
  • Text API rate limits list 500 RPM and 20,000,000 TPM for M2.7 and M2.7-highspeed.
  • MiniMax provides OpenAI-compatible and Anthropic-compatible integration paths.

As with prior MiniMax entries, maxOutput is stored as the documented total-token ceiling for repo comparability rather than a distinct output-only cap.

Pricing & Access

MiniMax’s current pay-as-you-go pricing lists:

  • MiniMax-M2.7 input: $0.30 per 1M tokens.
  • MiniMax-M2.7 output: $1.20 per 1M tokens.
  • Prompt cache read: $0.06 per 1M tokens.
  • Prompt cache write: $0.375 per 1M tokens.

M2.7-highspeed doubles token prices to 0.60inputand0.60 input and 2.40 output per 1M tokens while targeting faster inference. MiniMax also offers subscription-style token plans with request quotas for standard and high-speed M2.7 routes.

Best Use Cases

Choose MiniMax M2.7 for cost-sensitive coding agents, office workflow automation, long-context document work, and Chinese-English productivity assistants where you want hosted API economics closer to DeepSeek than to premium Opus or GPT tiers.

It is less ideal when governance, documentation maturity, western enterprise procurement, or a broad app ecosystem matter more than price-performance.

Comparisons

  • MiniMax M2.5 (MiniMax): Older M-series production tier; M2.7 is now the better default for new MiniMax evaluations.
  • DeepSeek V4 (DeepSeek): Similar low-cost Chinese API alternative with broader 1M-context positioning; M2.7 leans more explicitly into agentic productivity and office workflows.
  • Kimi K2.6 (Moonshot AI): Open-weight agentic coding peer; MiniMax M2.7 is proprietary hosted infrastructure with very aggressive pay-as-you-go pricing.