BenchmarksSWE-benchCost OptimizationMedium
How a $0.30 Model Beat Claude Opus on SWE-bench
MiniMax M2.5 augmented with PRAT-powered context via the Xanther Context Engine achieved 78.2% on SWE-bench Verified — surpassing Claude 4.5 Opus at 76.8%, at a fraction of the cost.
Xanther Team·Read →