humans& launches Ampersand, a 'collaborator' model it says beats frontier models
gharik · x · 2026-10-24
Startup humans& introduced Ampersand, billed as the first model built for collective goals — a collaborator trained to help people achieve individual and collective aims. In multi-month simulations, users working with Ampersand accomplished more than with Fable and other frontier models. Available to everyone via API.
The company is also open-sourcing its stack: contributions to FlashInfer, SGLang, and Miles, with its custom kernel powering Ampersand's large-scale RL training and Boba serving.
More from Models
- Kimi k3.1 pro aces the pelican-bike test in 45 minutes; author calls benchmark saturated — ChrisGPT · 2026-10-24
- urllib3 maintainer rejected by OpenAI's Cyber Verification Program: form is offensive-researcher only — matthew_d_green · 2026-10-24
- Mubit open-sources Banyan Sapling, a verification model with calibrated probabilities (Apache 2.0) — testingcatalog · 2026-10-24
- 105 planted bugs tested: Kimi K3.1 Pro scores 36, loses on cost-effectiveness — PawelHuryn · 2026-10-24
- Kimi K3.1 isn't an Opus-class model, says investor ahead of detailed benchmarks — bindureddy · 2026-10-24
- Baseten: Mirendil-Base-Large posts lowest perplexity on unseen private code — baseten · 2026-10-24