GPT-6 Sol reportedly worse than 5.6 Sol on DeepSWE and computer use; mocked as rebranded Terra

kristoph · x · 2026-09-23

An unverified claim says GPT-6 Sol performs worse than 5.6 Sol on DeepSWE, computer use and research debugging, with no gains in cyber — cheaper and more efficient, but not smarter. Kristoph quips that they basically rebranded Terra.

Related event: GPT-6 Sol/Luna reviews: cheaper and less hallucinatory, but trail predecessors on some benchmarks(16 posts)→

Original post →

More from Models

Models channel →