Vibeleaderboard
← All Intel
Intel / article

DeepSeek V4 Pro 0813 vs GPT-5.6 Sol on DeepSWE: Cost, Coding, and Routing

Source
together.ai
Date

Context

The cheapest coding attempt can also take longer. In Together AI's experiment on 113 DeepSWE tasks, DeepSeek V4 Pro 0813 cost $0.24 per run and took a median 35 minutes. OpenAI's GPT-5.6 Sol cost $8.37 and took 17 minutes. Sol also had the higher official single-attempt solve rate, 72.7% versus 62.8%. Both models ran at maximum effort with four trials per task. Experimental results.

The comparison separates two costs a team might otherwise conflate: the provider bill and the time spent waiting for a patch. In this setup, the more expensive model bought faster completion and better first-attempt results; the cheaper model suited work that could tolerate longer runs.

Read the source www.together.ai
Recommended reads
Comments

Checking sign-in…

Loading comments…