# MiniMax ships M3 — open-weight frontier claims, weights to follow

> MiniMax launched M3 on 1 June 2026, an open-weight frontier coding model, 1M context.

*Cheap, capable, and not quite downloadable yet. The brief.*

By The FeaturedDaily Desk · FeaturedDaily
Canonical: https://featureddaily.com/news/minimax-m3-launch-brief

> **Key:** **The one-liner:** MiniMax announced a frontier-class open-weight model and bargain pricing — then asked you to wait about ten days for the weights.

**What happened.** On 1 June, MiniMax released **M3**, billed as the first open-weight model to combine frontier coding, a 1M-token context and native image/video understanding, built on its new MiniMax Sparse Attention. It reports 59.0% on SWE-Bench Pro — ahead of GPT-5.5 and Gemini 3.1 Pro (MiniMax's own figures).

**The money.** Launch pricing of roughly $0.30 input / $1.20 output per million tokens (promotional) lands at about a tenth of the closed frontier — the genuinely disruptive part.

> **Note:** **What to watch.** The weights weren't published at launch (promised on Hugging Face within ~10 days), so the benchmarks couldn't be independently verified, and ARC-AGI-2 sits under 12%. Strong on coding, unproven on reasoning. Hong Kong shares swung +5% to -12% the same day.

## Key takeaways

- What: MiniMax M3, pitched as the first open-weight model fusing frontier coding, 1M context and multimodality.
- Price: ~$0.30/$1.20 per million tokens (promotional) — about a tenth of Opus and GPT-5.5.
- Catch: weights not released at launch (promised on Hugging Face within ~10 days); benchmarks vendor-run.
- Gap: under 12% on ARC-AGI-2 — behind US labs on abstract reasoning.

## FAQ

### What is MiniMax M3 in one line?
A Shanghai-built model launched 1 June 2026, pitched as the first open-weight system to combine frontier coding, 1M-token context and native multimodality — at roughly a tenth of the price of Opus or GPT-5.5.

### Are the weights out?
Not at launch — MiniMax said they'd reach Hugging Face within about ten days. Until then the benchmark claims are vendor-reported and not independently verifiable.