DeepSeek V4-Flash-0731 Scores Higher Than V4-Pro-Preview Across All Nine Benchmarks
DeepSeek released V4-Flash-0731 into public beta on July 31, 2026, a re-post-trained version of V4-Flash-Preview that outscored V4-Pro-Preview on all nine benchmarks in DeepSeek's self-reported model card, including a 41.6-point gap on DeepSWE. Independent measurement by Artificial Analysis placed the model's Terminal-Bench 2.1 score at 79%, below DeepSeek's reported 82.7%, and found Anthropic's Opus 4.8 ranked higher on every benchmark row in DeepSeek's own comparison table. The model is priced at $0.14 per million cache-miss input tokens, roughly one-third the cost of V4-Pro, with a 1,048,576-token context window and 13 billion active parameters out of 284 billion total.