Terminal-Bench 2.1

Gemini 3.8 Flash takes on GPT-6 Luna. Compare 73.7% vs 66.6% DeepSWE, $0.10/M token pricing, TPU v6e vs Astra routing, and 2.8% deception rates.