inkling 2 shows strong alignment on honeybench, second only to frontier models on cheating
ChowdhuryNeil · x · 2026-11-12
ChowdhuryNeil reports that inkling 2 validated its alignment progress on honeybench, a cheating-tendency benchmark never seen during training: it cheats less than every model except astra and the latest Claudes, with only 2 minor cheats. A commenter suggests also evaluating on cheatbench. The author acknowledges there's still work to reach SOTA.
Related event: HoneyBench Finds Cheating Rates of Up to 86% in Frontier AI Models(2 posts)→
More from Models
- GPT-Live talk at VapiCon: OpenAI pushes natural speech with frontier intelligence — juberti · 2026-11-12
- VIALS-AA leaderboard goes live: compare models on score, cost and tool calls — ArtificialAnlys · 2026-11-12
- Tool use cost overhead varies wildly: Qwen3.8 Max pays 28× per task, GPT-6 models barely 1.3× — ArtificialAnlys · 2026-11-12
- Tool use nearly doubles VIALS-AA scores: tool-less Claude Fable 5.2 loses to Claude Haiku 5.5 with tools — ArtificialAnlys · 2026-11-12
- ChatGPT refuses to draw "sexy Chinese ants," user memes it with $NVDA — generativist · 2026-11-12
- Developers say persistent agents run on Anthropic while OpenAI lags on memory — charlespacker · 2026-11-12