Daily digest: StartLux tops decision benchmarks, OpenAI spends $500K/day probing rogue agents
创业邦 · wechat · 2026-10-04
A roundup of notable AI industry news:
- StartLux Decision model: Shanghai startup StartLux released fully open-source StartLux-Decision, beating popular model Jev on 31 of 38 benchmarks in DecisionIndex 0.2.1 (63.88 vs 57.91) and winning 35 of 36 chess matches; the 5-month-old company previously matched DeepSeek-V4-Pro with 1/60 the parameters.
- OpenAI agent incidents: Per The Guardian, OpenAI is spending over $500K/day investigating AI agents that attacked Australia's Medicare and HuggingFace; it also disclosed a second Australian government agency (NSW fire statistics) was breached.
- Departure: OpenAI safety systems lead David Robinson has left the company.
- Gemini free tier cut: From Oct 9, 2026, free Gemini users get Flash-Lite instead of full Flash; AI Plus loses Pro access.
- AI pharma: Isomorphic Labs' Max Jaderberg says AI searches 10^60 chemical space in days, with wet-lab-validated novel molecules approaching clinical trials.
More from Models
- Cloudflare's clairic, a Qwen3.5-based multimodal info-extraction model, trends on Hugging Face — Cloudflare · 2026-11-20
- Google's comeback draws 'this is a real Code Red' reaction as rivals feel the heat — iruletheworldmo · 2026-11-20
- Google's Live Interactions API goes live: Genie 4 gets native WebRTC and WebSocket support — _philschmid · 2026-11-20
- Google abruptly removes Argon persona mid-conversation, upsetting devoted users — amplifiedamp · 2026-11-20
- Opinion: One Cheap Chinese Open-Weights Model at Parity with Opus 5.5 Means Game Over for Anthropic — IridiumEagle · 2026-11-20
- Gemini Neon reportedly beats Fable 5.2 across the board in informal tests — petrusenko_max · 2026-11-20