MiniMax-M1-40k

Released
2025-06-16
Last refreshed
2026-10-09
Status
Researched today
Open sourceCommercial use: permittedCodingRAGAgentsLong contextClassificationJSON / Tool use

MiniMax-M1-40k is available now for coding, rag, and agents with open-source and 1m context; evaluate it while provider pricing coverage matures.

MiniMax releases · 9 in the last 12 monthsChangelog →
Specifications
Released
2025-06-16
Context
1m
Parameters
456B total, 45.9B active (MoE)
Architecture
Mixture of Experts
Specialization
general
Openness
Open source
License
Apache 2.0OSI-approvedCommercial use: permitted
Weights
Available
Code
Unknown
Training
Pretrained
Created by

Developing AI for gaming and entertainment.

Minhang, Shanghai, China
Founded 2021
Website
Pricing

No tracked provider token pricing is available yet.

About

MiniMax-M1-40k is the 40K-thinking-budget version of MiniMax-M1, MiniMax's open-weight reasoning model released on 16 June 2025. It shares M1's hybrid mixture-of-experts design with lightning attention, about 456 billion total and 45.9 billion active parameters per token, and a native context window of 1 million tokens; the 40K figure is how much the model was trained to spend on reasoning, not its context length. MiniMax trained it with large-scale reinforcement learning using its CISPO method and released it alongside the longer-thinking MiniMax-M1-80k.

Provider price ladder

No tracked provider token pricing is available for this model yet.

Capabilities

ReasoningJSON / Tool use

Benchmark peer barsfor Coding

Benchmark scores(9)

Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.
BenchmarkScoreVersionEvaluationSource
AIME 202483.3AIME 2024, 40K thinking budget (accuracy)Observed 2026-10-09—Source
AIME 202574.6AIME 2025, 40K thinking budget (accuracy)Observed 2026-10-09—Source
MATH-50096.0MATH-500 (accuracy)Observed 2026-10-09—Source
LiveCodeBench62.3LiveCodeBench 24/8~25/5 (accuracy)Observed 2026-10-09—Source
Google-Proof Q&A69.2diamondObserved 2026-10-09—Source
Humanity's Last Exam7.2HLE text-only, no tools (accuracy)Observed 2026-10-09—Source
MMLU PRO80.6MMLU-Pro (accuracy)Observed 2026-10-09—Source
SWE-bench Verified55.6SWE-bench VerifiedObserved 2026-10-09—Source
MultiChallenge44.7MultiChallenge (accuracy)Observed 2026-10-09—Source

Migration checks

No linked migration route is available for this model yet.

API versions

MiniMax-M1-40kMiniMaxAI/MiniMax-M1-40kMiniMaxAI/MiniMax-M1-40k-hfminimax-m1-40k