MiniMax launches M3.1 Flash, a cheap agent and coding model, but independent tests flag a high hallucination rate
Announced · Published · By Contractor AI Shop
MiniMax released M3.1 Flash through its API, M Plan and MiniMax Code at $0.15 per million input tokens and $0.60 per million output tokens. Artificial Analysis scored it 42 on its Intelligence Index but measured a 93% hallucination rate. MiniMax says weights come later.
Key facts
- MiniMax lists flat pricing of $0.15 input, $0.60 output and $0.015 cached input per million tokens, and says overall usage cost is 70% lower than its predecessor. MiniMax
- VentureBeat reports Artificial Analysis scored M3.1 Flash 42 on its Intelligence Index, versus 45 for DeepSeek V4.1 Flash-1017. VentureBeat (Carl Franzen)
- Artificial Analysis found a 93% hallucination rate on its AA-Omniscience test: when the model didn't answer correctly, it almost always gave a wrong answer instead of saying it didn't know. VentureBeat (Carl Franzen)
- SiliconANGLE reports MiniMax's developer guide for the preview lists 428 billion total parameters, so this is not a model for shop hardware. SiliconANGLE (Duncan Riley)
Our take: what it means for contractors
The price war on cheap agent models keeps going, but the part that matters for contractors is the hallucination result. A model that confidently answers when it doesn't know shouldn't fill in spec, code or bid facts without retrieval from your documents and a human check.
What we don't know yet
- When the weights ship and under what license (VentureBeat says Oct 23; MiniMax only says later)
- Memory needed to run a quantized build
- How the China-hosted API handles uploaded documents
Sources
- MiniMax · MiniMax M3.1 Flash: Redefining Agent Performance and Cost · 2026-10-19 · Primary
- VentureBeat (Carl Franzen) · MiniMax's new M3.1 Flash beats DeepSeek and Opus 5 on a key business benchmark at lower cost, but with a high hallucination rate · 2026-10-19 · Secondary
- SiliconANGLE (Duncan Riley) · MiniMax launches M3.1 Flash model, says it cuts AI agent costs by 70% · 2026-10-19 · Secondary