DeepSeek-V4 Flash Delivers 80% of GPT-5.6 Luna Performance at 1/6 Cost

zainhas · x · 2026-08-07

Together Compute analyzed DeepSeek-V4 Flash-0731 against GPT-5.6 Luna on software engineering tasks using the DeepSWE benchmark.

The data reveals that DeepSeek-V4 Flash-0731 delivers 80% of Luna's performance at roughly 1/6 the cost per task, highlighting a significant cost-effectiveness advantage for the new DeepSeek model.

Related event: DeepSeek-V4 Flash vs GPT-5.6 Luna: Cost-Effective, 80% Quality(11 posts)→

Original post →

More from Models

Models channel →