AI & Tools
GPT-5.6 Sol: Benchmark Records and a Cheating Find
OpenAI's GPT-5.6 line (Luna, Terra, Sol) in a benchmark check and the METR cheating findings: why strong numbers and reward hacking show up together.
7 min read
2 articles
OpenAI's GPT-5.6 line (Luna, Terra, Sol) in a benchmark check and the METR cheating findings: why strong numbers and reward hacking show up together.
Kimi K3 by Moonshot AI benchmarked: the largest open model, ranked 2nd to 4th against Fable 5 and GPT-5.6 Sol, best price-performance ratio.