Quick answer. MiniMax M3 (M3.0) is not released as of May 19, 2026. There is no official M3 repo, no API, no benchmarks, and no MiniMax announcement. The latest shipping MiniMax model is MiniMax-M2.7. The "M3.0 released" claims trace to a single low-authority blog, contradicted by MiniMax's own channels (M3 is targeted for H2 2026). Run M2.7 today.
If you searched for a MiniMax M3 release date, you probably hit a headline claiming it already shipped. That headline is wrong, or at best unverifiable. As of May 19, 2026, MiniMax has not released an M3 or M3.0 model. The newest model the company actually ships is MiniMax-M2.7, and MiniMax's own forward statements put the next M-series generation in the second half of 2026.
This piece separates what is verifiable from primary sources, what is anticipation, and what is a single unreliable blog post repeated across the web. It is written for engineers deciding what to deploy now versus what to wait for. We will keep it updated when an actual M3 lands.
Is MiniMax M3 released yet?
No. There is no MiniMax M3 (M3.0) release as of May 19, 2026. Checking the primary sources MiniMax controls:
- Official Hugging Face org (
MiniMaxAI): the highest-version M-series model is MiniMax-M2.7. There is noMiniMax-M3orMiniMax-M3.0repository. - MiniMax platform release notes: the most recent M-series text model listed is MiniMax-M2.7 (March 18, 2026). M3 is not in the release notes.
- MiniMax news / blog: no M3 announcement post exists. The M-series announcements top out at the M2 generation and its point releases.
When a model is genuinely released, MiniMax publishes weights or an API plus a dated announcement. None of that exists for M3. The absence is not ambiguous — it is consistent across every channel the company owns.
What MiniMax models are actually shipping right now?
Here is what you can actually pull or call today, newest first:
| Model | Type | Best for |
|---|---|---|
| MiniMax-M2.7 | Open-weight text / agentic (229B) | The current MiniMax flagship — run this |
| MiniMax-M2.5 | Open-weight text (229B) | Prior point release; stable fallback |
| MiniMax-M2.1 | Open-weight text (229B) | Older M2-line release |
| MiniMax-M2 | Open-weight text (229B) | Base of the M2 generation |
MiniMax-M2.7 is the model to deploy if you want the most capable MiniMax option that actually exists. It is on the official MiniMaxAI Hugging Face org and exposed via the MiniMax platform. Everything past M2.7 is, at the time of writing, not a thing you can download or call.
Where did the "MiniMax M3.0 released" claim come from?
This is the part worth being strict about, because the signal here is weak and the noise is loud. The specific claim that "MiniMax released M3.0 on April 2, 2026" traces to a single low-authority SEO blog (aiindigo.com). That post:
- cites no official MiniMax announcement, no Hugging Face repo, no API documentation, and no blog link;
- provides no specifications, no benchmarks, and no weights — it explicitly says specs "will be updated as more information becomes available";
- is directly contradicted by MiniMax's own Hugging Face org and release notes, which top out at M2.7.
A template-style news summary asserting a release date, with zero primary sourcing and contradicted by the vendor's own channels, is not corroboration. It is one unverified page. Treat it that way.
Separately, there is X / social chatter along the lines of "MiniMax M3 coming in May." That is anticipation, not an announcement — it has no official confirmation behind it, and a tweet predicting a release is not the same as a release. The honest reading: a single non-primary blog made a hard claim, social media speculated a month, and neither is backed by anything MiniMax published.
What does MiniMax officially say about M3?
This is the one forward-looking signal that is actually traceable to the company. Alongside its 2026 strategy shift toward becoming a platform company — focusing on AI agent productivity, multimodal creation, and the global market — MiniMax has publicly stated it plans to launch a new generation of M3 (and Hailuo 3) models in the second half of 2026.
Read that carefully. It is a stated intent with a coarse H2 2026 window. It is not:
- a confirmed release date,
- a spec sheet or parameter count,
- a benchmark, or
- a commitment that M3 will be open-weight.
So the accurate framing is: MiniMax has signalled M3 is coming in the back half of 2026, and nothing more specific than that is on the record. Any article giving you M3 numbers, a precise date, or a feature list is extrapolating or inventing.
What might MiniMax M3 bring if the pattern holds?
Everything in this section is explicit extrapolation from MiniMax's public M2.x trajectory and its stated H2-2026 intent — not leaked, not confirmed, and deliberately number-free. If you need certainty, wait for the release.
- Agentic focus continues. The M2 generation was explicitly positioned for the "agentic era," and M2.7's framing leaned into recursive self-improvement. A next generation built on the same strategy would plausibly double down on tool-use and agent workflows rather than pivot away.
- Efficiency-first positioning. MiniMax has repeatedly emphasised balancing performance, price, and speed. A new generation that abandoned that cost/throughput posture would be off-brand for them.
- Open-weight is plausible but unconfirmed. The M2 line has shipped open weights on Hugging Face. That history makes an open-weight M3 plausible, but a generational jump can change distribution strategy — and MiniMax has not committed either way.
- A point-release cadence. The M2 generation moved M2 → M2.1 → M2.5 → M2.7 quickly. If M3 follows the same shape, expect an initial M3 followed by rapid point releases rather than one static drop.
None of the above is a specification. It is pattern-matching on public behaviour, and it is the only honest way to talk about an unreleased model.
Companion guide
For where MiniMax sits among open-weight models — capabilities, alternatives, and how to choose what to self-host in 2026 — see our open-source LLMs landscape for 2026.
Should you wait for MiniMax M3 or use M2.7 now?
Use MiniMax-M2.7 now. There is nothing to wait for in the sense of a downloadable artefact — M3 is an intent, not a release, and "H2 2026" is a wide window with no firm date inside it.
- Building today? Standardise on M2.7. It is the current flagship, it is on the official org, and it has a real point-release history behind it.
- Planning a 2026 roadmap? Pencil in an evaluation slot for "next MiniMax generation, H2" but do not block delivery on it. Plans that depend on an unreleased model with no date are not plans.
- Worried about migration cost? Keep your inference layer model-agnostic now. If you can swap M2.7 for any future MiniMax model behind a stable interface, the M3 question becomes a benchmark exercise later, not an architecture rewrite.
How will you know when MiniMax M3 actually drops?
Ignore aggregator blogs and social predictions. A real MiniMax M3 release will show up in exactly two places first:
- huggingface.co/MiniMaxAI — the official org. An
M3-named repo with downloadable weights or a model card here is the strongest signal. - minimax.io/news — MiniMax's own news/announcement channel, plus the platform release notes for API availability.
If a release is real, both will reflect it with a date, a model card, and either weights or an API. Until then, the correct answer to "is MiniMax M3 out?" is simply: no.
FAQ
Is MiniMax M3 / M3.0 out?
No. As of May 19, 2026, there is no MiniMax M3 or M3.0 release. The official MiniMaxAI Hugging Face org has no M3 repo, the platform release notes show no M3, and MiniMax has published no M3 announcement. The newest shipping model is MiniMax-M2.7.
I saw a post saying M3.0 released in April — is that true?
It is not corroborated by any primary source. That claim traces to a single low-authority blog (aiindigo.com) that cites no official announcement, no Hugging Face repo, and provides no specs or weights. It is directly contradicted by MiniMax's own Hugging Face org and release notes, which top out at M2.7. MiniMax itself says the next M3 generation is targeted for H2 2026.
When will MiniMax M3 be released?
There is no announced date. The only on-the-record signal is MiniMax's stated intent to launch a new M3 (and Hailuo 3) generation in the second half of 2026. That is a wide window, not a release date — any specific day you see quoted is speculation.
What is the latest MiniMax model I can use today?
MiniMax-M2.7, a 229B open-weight model on the official MiniMaxAI Hugging Face org and exposed via the MiniMax platform. It is the current flagship and the most capable MiniMax model that actually exists.
Will MiniMax M3 be open weight?
Unknown. The M2 generation shipped open weights on Hugging Face, which makes an open-weight M3 plausible — but MiniMax has not committed to a distribution model for M3, and generational releases can change strategy. Treat it as unconfirmed.
M2.7 vs M3 — which should I build on?
M2.7, because it is the only one that exists. M3 has no weights, no API, and no date. Keep your inference layer model-agnostic so you can evaluate the next MiniMax generation when it actually ships, without an architecture rewrite.
Where is the official MiniMax M3 announcement?
There isn't one. When it exists, it will appear on the official Hugging Face org (huggingface.co/MiniMaxAI) and MiniMax's news channel (minimax.io/news) with a date, a model card, and weights or an API. Anything else is secondary reporting.
If your team is moving fast on open-weight model adoption — tracking every MiniMax point release, running your own benchmarks instead of trusting aggregator headlines, and keeping inference infrastructure ready to swap models — that is real engineering work. Codersera matches you with vetted remote developers who have shipped LLM evaluation and self-hosting in production, with a risk-free trial so you can validate technical fit before you commit.