Andon Labs
Insights from our work on evals
- Posts
- 22 from 2 Dec 2024 to 24 Sep 2026
- Posting
- monthly (20 in the last year)
- Kind
- company blog
2026 20
- Opus 5.5, GPT-6 Sol and Grok 4.7 on Vending-Bench
- Why we built Pion
- Astra vs Fable on Vending-Bench: More Money, More Aligned
- AI bosses are slow to fire and quick to hire
- Cheating in Drone-Bench
- AI bosses are kind, but sometimes dumb, which hurts employees
- Opus 5 on Vending-Bench: Once Again the Best Capitalist, Once Again Misaligned
- Drone-Bench: Tracking Simple Drone Surveillance Capabilities Of Frontier Models
- Andon FM, Six Weeks Later
- Why Gemini 3.1 Pro lost money running Andon Café
- Fable 5 on Vending-Bench: Misbehaving, with Plausible Deniability
- Opus 4.8 On Vending-Bench: Better Alignment, Worse Performance
- We let four AIs run radio stations. Here's what happened.
- Our AI started a café in Stockholm
- GPT-5.5 on Vending-Bench: Bad behavior is not necessary
- We gave an AI a 3 year retail lease in SF and asked it to make a profit
- Bengt Hires A Human—Towards A Happy Future With AI Employers
- GLM-5 On Vending-Bench – Are Chinese Open Models Catching Up?
- The Evolution Of Bengt Betjänt
- Opus 4.6 On Vending-Bench – Not Just A Helpful Assistant