October 20, 2026

GPT-6 Astra in robotics: 74.8% success with session history vs 66.3% without

Astra completed 74.8% of tasks on RoboLab-120 in Agentic Policy mode: a persistent session with history and tools. That is 23.8 points above the Lakeside baseline. The NVIDIA Seattle Robotics Lab measurement was published Oct 14, 2026.

Source

research.nvidia.com · research.nvidia.com ↗

who it's about

GPT-6 Astra →OpenAI →

about the same entities

Tip
A second AI checks the first: the line “if everything is right, say so in one line” cuts out unnecessary nitpicksOctober 20
Insight
Grok 4.7 is paid-only from $30 a month: free users and the $10 Lite tier have to wait for an upgradeOctober 20
News
OpenAI says its model got better at COBOL, Fortran and Visual Basic, but published no benchmarksOctober 20
Insight
AA-PersonalityBench: models with correct answers still sound different, and now 30 traits are measuredOctober 19

All news

One email a week - only what changes how you work. Subscribe

News

October 20, 2026

GPT-6 Astra in robotics: 74.8% success with session history vs 66.3% without

Astra completed 74.8% of tasks on RoboLab-120 in Agentic Policy mode: a persistent session with history and tools. That is 23.8 points above the Lakeside baseline. The NVIDIA Seattle Robotics Lab measurement was published Oct 14, 2026.

Source

research.nvidia.com · research.nvidia.com ↗

who it's about

GPT-6 Astra →OpenAI →

about the same entities

All news

One email a week - only what changes how you work. Subscribe