Making AI Headlines · Page 7
MiniMax Makes the Benchmark Table Bring a Calculator
· New York time
MiniMax's M3.1 Flash scored 42 on the Artificial Analysis Intelligence Index, a 13-point improvement over its predecessor, at half the cost. It also reached 73.7 percent on the DeepSWE v1.1 coding benchmark for under 30 cents per task, matching DeepSeek V4.1 Flash 1017 on reported performance and price. The results add pressure to global competition over AI cost and capability. For American developers comparing models, the practical attraction is not national scoreboard theater. It is the possibility of completing a particular workload with less spending. A benchmark is useful evidence, but its relevance depends on how closely the test resembles the work someone actually needs done. Lower prices can widen access and make experimentation cheaper. They also push providers to explain what customers receive for a premium beyond a more prestigious name in the model selector. The industry's old magic trick was making software produce an answer. The newer trick is making the invoice small enough that someone can afford to ask again. Still, a benchmark table is not a universal purchasing decision. Performance, reliability, and fit remain part of the assessment, even when the cost column is doing its best to steal the meeting.
Read this story in the full newspaper →Satire. Not factual reporting. Read the original reporting from the linked publishers.