RoboVista benchmark: Gemini-4 Argon hits 81.9% on robotics understanding
berkeley_ai · x · 2026-10-16
Researchers from UC Berkeley, Google DeepMind, and Princeton unveiled RoboVista at RSS 2026 — a systematic benchmark evaluating how well today's VLMs understand diverse, real-world robotics applications, with website, dataset, and paper released.
Latest scores:
- Gemini-4 Argon: 81.9% at just $2.54 total eval cost, the leader
- GPT-6 Astra: 75.3% ($6.08)
- Claude Fable 5.1: 64.1% ($7.94)
Compared to Gemini 3 Flash, Gemini-4 Argon gains +12.9 points overall and improves in every domain: Driving 75→100 (+25.0), Industry 58→76 (+18.3), Agriculture 70→87 (+17.0), plus gains in open data, surgery, and home scenarios.
More from Models
- OpenAI stealth-testing mysterious internal model 'Aeon' in Codex desktop app and Dots — cedric_chee · 2026-10-16
- Argon edges out astra as daily driver, but Claude still wins heavy coding work — iruletheworldmo · 2026-10-16
- NVIDIA ships FP4-quantized Qwen3.8-Flash for local use, trending on Hugging Face — nvidia · 2026-10-16
- Dev reports decision model jev beats gpt-6-luna on quality in real training pipeline — genmon · 2026-10-16
- Google won't publish Argon 4 limits in AI Pro—users compare experiences — Consistent-Text-2662 · 2026-10-16
- Reddit user puzzled why so many prefer Claude over 'much better' ChatGPT — -ButterflyEffect- · 2026-10-16