Calibrated on 193 published benchmarks

Coding models you can run

Models trained for code, agents and tool use, graded for your hardware.