GLM-5.3 matches frontier models on DeepSWE at fraction of cost

Together AI ran 904 DeepSWE rollouts: GLM-5.3 hits 69.0% pass@1 vs. GPT-5.6 Sol's 72.7% and Claude Fable 5's 69.7%, but costs $3.99 per rollout vs. $8.37 and $21.63. A GLM-first cascade solves 85.9% of tasks at $6.61 each.