AI & ML · Jul 31
DeepSeek V4 Flash 0731 hits 50 and makes GPT-5.6 Luna look pricey
Artificial Analysis says DeepSeek gained 10 Intelligence Index points while undercutting a comparable Luna task cost through aggressive cache pricing.
- Compare models by total task cost, not only benchmark score, especially for repeated long-context workloads.
- Test cache hit rates on your own prompts because DeepSeek’s pricing edge depends heavily on cache economics.
- Watch for full weights and independent latency data before committing production workloads.