gpt-6IMAGE
GPT-6 Sol is worse than 5.6 Sol on several benchmarks, leak claims
A post with an attached image claims Sol regresses on DeepSWE, computer use and research debugging, shows no gain in cyber, and is only cheaper and more efficient.
NIK
@ns123abcArchived 2026.09.23Originally posted 2026.09.23




NIK @ns123abcOriginally posted 2026.09.23
🚨 BREAKING: GPT-6 Sol confirmed WORSE than 5.6 Sol on DeepSWE, computer use and research debugging, and no better at cyber it's cheaper and more efficient, but not a smarter model
x.com/ns123abc/status/2102474380332654608Archived 2026.09.23 10:20