Conspiracy theory claims MiniMax sniped Artificial Analysis queries to train M3.1 Flash

AI researcher 'Binkg' alleges MiniMax saw Artificial Analysis' AA-Omniscience prompts through its inference servers during the M3 evaluation and then trained M3.1 Flash on them. The 428B-parameter model (23B active) scored 42 on the AA Intelligence Index, 13 points above M3. Artificial Analysis and MiniMax have not commented.

• M3.1 Flash scored 42 on the AA Intelligence Index vs 29 for M3
• Only 600 of roughly 6,000 AA-Omniscience questions are public
• M3 launched via API on June 1, AA benchmarks on June 4, weights on June 12
• Pricing: $0.15 per 1M input and $0.60 per 1M output tokens