DeepSeek V4 Pro 0813 vs Claude Fable 5 on DeepSWE: Cost, Coding, and Routing
Together AI released performance benchmarks for DeepSeek V4 Pro 0813 and Claude Fable 5 on the DeepSWE coding benchmark. The data establishes a cost-performance trade-off where Claude Fable 5 leads in pass@1 accuracy while DeepSeek V4 Pro 0813 excels in pass@4 and cascade routing efficiency.
Verified State Diff
Impact & Verification Analysis
Software engineers, AI infrastructure architects, and enterprise developers utilizing LLMs for automated coding tasks.
It provides a quantitative framework for optimizing LLM routing strategies, allowing developers to significantly reduce operational costs by cascading between high-cost, high-accuracy models and cost-efficient alternatives without sacrificing overall system performance.
Full Fact Overview
The analysis evaluates two specific models, DeepSeek V4 Pro 0813 and Claude Fable 5, across 904 rollouts on the DeepSWE benchmark. Claude Fable 5 demonstrates superior pass@1 performance but at a 90x higher cost compared to DeepSeek V4 Pro 0813. Conversely, DeepSeek V4 Pro 0813 achieves higher pass@4 success rates. The implementation of a Pro-first cascade routing strategy yields an 82.7% success rate, suggesting an architectural optimization for balancing high-accuracy coding tasks with cost-effective model selection.