A comparison of DeepSeek V4 Pro 0813 and Claude Fable 5 on the DeepSWE benchmark reveals that a routing strategy using both models solves 82.7% of tasks at $8.28 each, outperforming Fable alone (69.7%) while being significantly cheaper.
- Claude Fable 5 leads in single-attempt accuracy (69.7% pass@1) and excels in Rust (85%) and serialization-heavy work, but costs $21.63 per rollout, roughly 90 times more than DeepSeek V4 Pro 0813.
- DeepSeek V4 Pro 0813 is cheaper ($0.24 per rollout) and achieves higher accuracy with multiple attempts (88.5% pass@4 vs Fable's 84.1%), making it the better value for high-volume or retry-tolerant work.
- The models show low per-task correlation (0.39), covering 107 of 113 tasks between them, which justifies a cascade approach: running Pro first and escalating to Fable only when necessary.
The analysis suggests that Fable 5 is best used as a selective escalation for specific domains like Rust rather than a default model, while DeepSeek V4 Pro 0813 offers near-Fable accuracy at a fraction of the cost.