We analyzed DeepSeek-V4 Flash-0731 vs. GPT-5.6 Luna on software engineering tasks using DeepSWE. DeepSeek-V4 Flash-0731 delivers 80% of Luna…

We analyzed DeepSeek-V4 Flash-0731 vs. GPT-5.6 Luna on software engineering tasks using DeepSWE. DeepSeek-V4 Flash-0731 delivers 80% of Luna’s performance at roughly 1/6 the cost. More insights in the thread! 👇 Deepdive: DeepSeek-V4 Flash 0731 [max] vs. GPT 5.6 Luna [max] on software engineering/DeepSWE tasks. > DeepSeek flash is 1/6th the cost of Luna per task. > V4 flash 0731 is 80% the quality of Luna DSv4 flash is insane value for money 🤯 full deep-dive 👇(1/n)🧵

Source: Together AI (X) | 2026-08-07