MiniMax M2.7 release date: March 18, 2026, and it is callable today.

Quick answer: the verified timeline.

MiniMax M2.7 was released on March 18, 2026. The date comes from MiniMax's own announcement, which shipped the model to its Agent product and API platform simultaneously, positioning it for agentic workflows, software engineering, and office tasks. The open weights followed on Hugging Face under a Modified-MIT license, with coverage of the open-weight release dating to April 2026; the exact day the checkpoint landed is less consistently reported than the model's release, so this page pins the month for the weights and the day for the model. So the release-date question has a closed answer: the model is out, the API id is stable, and the weights are public. The open questions are what shipped exactly and what is known about the next release, both covered below.

The M2 family history: four releases in five months.

MiniMax's M2 line moved faster than almost any peer family, and MiniMax itself noted the pace exceeded its own expectations. MiniMax-M2 arrived on October 27, 2025 as the line's foundation, an efficiency-first MoE aimed at coding and agent work. M2.1 followed on December 23, 2025, with a state-of-the-art claim on SWE-bench Multilingual (72.5) and strong application-development benchmarks. M2.5 landed on February 12, 2026, with MiniMax claiming top-class results across coding, agentic tool use, search, and office work (80.2 on SWE-Bench Verified and 76.3 on BrowseComp with context management among its reported figures). M2.7 closed the sequence on March 18, 2026, adding the launch's signature claim: an internal version participated in its own reinforcement-learning development. All recent ids remain available on MiniMax's platform, and the naming skipped around (no public M2.2 through M2.4, no M2.6), which is worth knowing before you conclude a missing version means missed news.

What shipped on March 18: verified vs vendor-claimed.

One catalog note for anyone budgeting long prompts: the MiniMax-M2.7 entry here lists a 1M context while the vendor's spec says roughly 200K. Budget against the vendor figure until an oversized test request on this route proves otherwise; the setup page linked below covers the test.

• Verified from vendor documentation: a sparse Mixture-of-Experts of about 230B total parameters with roughly 10B active per token, and a context window on the order of 200K tokens (196,608 in the spec).
• Verified from the release channel: day-one availability on MiniMax Agent and the API platform, with open weights on Hugging Face following under a Modified-MIT license.
• Vendor-reported benchmarks, treat accordingly: 56.2 on SWE-bench Pro, 57.0 on Terminal-Bench 2.0, 76.5 on SWE-bench Multilingual, 77.8 on BrowseComp.
• Vendor claim, unverifiable from outside: that an internal version of the model autonomously handled a meaningful share of the RL research workflow that produced it, running 100-plus optimization rounds. Memorable, load-bearing for marketing, not for your routing decision.
• License fine print with its own news cycle: the Modified-MIT terms require MiniMax's prior written authorization for commercial use of the weights, a restriction coverage highlighted in April 2026.

What is not confirmed.

No M2.8, M2.9, or M3 has been announced. Pages tracking an M3 status list rumor-level material only, and MiniMax has published no roadmap dates as of July 2026. The family's observed cadence, roughly a release every six to nine weeks through March, then a pause through the spring, supports no confident prediction in either direction; a pattern is not a date, and a pause is not a signal. The self-evolution claim also stays in the unconfirmed column as a matter of method: it describes MiniMax's internal development process, cannot be independently verified, and has no bearing on how the model behaves on your traffic. Evaluate M2.7 as a model, not as a story. The practical hedge against roadmap uncertainty is the usual one: integrate against the exact id MiniMax-M2.7 behind an endpoint that serves multiple families, and whatever MiniMax ships next becomes an A/B rather than an event you wait on.

How to call MiniMax M2.7 today.

The released model is addressable now through any OpenAI-compatible client: base URL https://api.apisrouter.com/v1, model MiniMax-M2.7 with its exact capitalization, one key that also covers the GLM, DeepSeek, Qwen, and Kimi ids you would compare it against. No MiniMax account or mainland onboarding is required through the gateway, and the starting credit covers the first requests while you evaluate.

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["APISROUTER_API_KEY"],
    base_url="https://api.apisrouter.com/v1",
)

resp = client.chat.completions.create(
    model="MiniMax-M2.7",   # released March 18, 2026
    messages=[{"role": "user", "content": "What changed between MiniMax M2.5 and M2.7?"}],
    max_tokens=2000,
)
print(resp.choices[0].message.content)

Pay-as-you-go · transparent per-model pricing

Selected models are priced below official list prices. Exact input, output, cache, and per-request prices are shown for each model.

Why M2 release dates keep confusing people.

Three patterns generate the noise. Skipped version numbers: the public sequence ran M2, M2.1, M2.5, M2.7, so searches for the missing versions surface speculation rather than releases, and readers infer unannounced models that never existed. Model-then-weights staging: the March 18 API release and the April open-weight release are both correct dates for different artifacts, and coverage of each reads as the release. And platform lag: third-party platforms listed M2.7 on their own schedules after launch, so launch-week answers to "is it out?" differed per endpoint. The reliable check beats all of it: list the models your endpoint serves and look for the exact id. If MiniMax-M2.7 appears in /v1/models, it is released for you; everything else is press-cycle detail.

curl -s https://api.apisrouter.com/v1/models \
  -H "Authorization: Bearer $APISROUTER_API_KEY" | grep -i minimax

FAQ