DeepSeek V4 Pro and GPT-5.6 Sol cascade solves 83% of DeepSWE tasks at $3.35

A comparison of DeepSeek V4 Pro 0813 and GPT-5.6 Sol on the DeepSWE benchmark demonstrates that running DeepSeek first and escalating to GPT-5.6 Sol upon test failure achieves an 83.0% pass rate at $3.35 per task.

• DeepSeek V4 Pro 0813 costs $0.24 per rollout, making it 35x cheaper than GPT-5.6 Sol's $8.37 price tag.
• While GPT-5.6 Sol leads single-attempt accuracy with a 72.7% pass@1 rate, DeepSeek V4 Pro 0813 surpasses it at four attempts with an 88.5% pass@4 score.
• The cascade strategy beats both models running alone and outperforms a perfect one-shot oracle router by achieving 83.0% coverage for less than half of GPT-5.6 Sol's per-task cost.
• GPT-5.6 Sol remains superior for latency-sensitive tasks, completing rollouts in 17 minutes compared to DeepSeek's 35 minutes, and wins six of eight task domains including Python and Go.

This routing approach allows users to leverage DeepSeek V4 Pro 0813 as a low-cost front-end that clears most workloads, reserving the expensive GPT-5.6 Sol only for difficult tasks that require its higher precision.