MiniMax has launched its latest model, MiniMax-M2.7, signaling a significant shift in the balance between high-level intelligence and operational affordability. By scoring a 50 on the Artificial Analysis Intelligence Index, the model now stands shoulder-to-shoulder with established competitors like GLM-5. What makes this achievement particularly striking is the price point; the M2.7 delivers this performance at less than one-third the cost of its rivals, maintaining the same aggressive pricing structure as its predecessor despite the performance leap.

The update focuses heavily on agentic tasks—scenarios where an AI must navigate complex, multi-step workflows to complete a goal—reaching an impressive Elo score of 1494. This leap is largely attributed to a refined approach to factual accuracy. The model has seen a dramatic reduction in hallucinations, which are instances where an AI confidently generates false information. By training the system to abstain from answering when it lacks sufficient data, MiniMax has brought its hallucination rate down to 34%, outperforming several larger frontier models.

While the model is more verbose, using approximately 55% more output tokens than the previous M2.5 version to reach its conclusions, its efficiency remains market-leading. It currently sits on the Pareto frontier, a technical state describing the best possible trade-off between intelligence and cost. For developers, this means accessing reasoning capabilities that previously required a much higher budget, though the model remains text-only and lacks multimodal features like image or audio processing at this stage.

MiniMax has launched its latest model, MiniMax-M2.7, signaling a significant shift in the balance between high-level intelligence and operational affordability. By scoring a 50 on the Artificial Analysis Intelligence Index, the model now stands shoulder-to-shoulder with established competitors like GLM-5. What makes this achievement particularly striking is the price point; the M2.7 delivers this performance at less than one-third the cost of its rivals, maintaining the same aggressive pricing structure as its predecessor despite the performance leap.

The update focuses heavily on agentic tasks—scenarios where an AI must navigate complex, multi-step workflows to complete a goal—reaching an impressive Elo score of 1494. This leap is largely attributed to a refined approach to factual accuracy. The model has seen a dramatic reduction in hallucinations, which are instances where an AI confidently generates false information. By training the system to abstain from answering when it lacks sufficient data, MiniMax has brought its hallucination rate down to 34%, outperforming several larger frontier models.

While the model is more verbose, using approximately 55% more output tokens than the previous M2.5 version to reach its conclusions, its efficiency remains market-leading. It currently sits on the Pareto frontier, a technical state describing the best possible trade-off between intelligence and cost. For developers, this means accessing reasoning capabilities that previously required a much higher budget, though the model remains text-only and lacks multimodal features like image or audio processing at this stage.