Debate: are open models already at GPT-5.6 Sol-level cyber capabilities?
AlexBarry4 · x · 2026-10-22
- Alex Barry pushes back on claims that open models already match GPT-5.6 Sol-level cyber capabilities: public data is slim, GLM 5.3 is closer to GPT-5.5, and DeepSeek V4.1 Flash sits in the same ballpark, but its 1017 update shipped without cyber eval results.
- He notes the only supporting evidence is a single vals datapoint with very different token usage and a narrow benchmark, and asks which models the claim refers to.