GPT-6 Astra completes driving course in benchmark

In a driving benchmark discussed on HN, GPT-6 Astra with Codex at medium setting completed the course on its second attempt at 100% progress in 5:22, using 6.6M tokens and $7.74. Claude Fable 5.1 reached only 45% and Grok 4.6 11%, showing a capability gap on embodied driving tasks.

• Sphinx launches Git hook that quizzes developers on AI code before push2026-10-18
• Graphicle Brings Code Graph and Memory to Coding Agents2026-10-14
• Third Moment Research runs 10,000 agents for 36+ hours2026-10-14
• Pensieve local indexing cuts coding-agent tool calls2026-10-12
• Asana cuts model costs 76x with GPT-6.1 Sol2026-10-09