Customers report separate GPT-6 Astra efficiency gains

Parallel says GPT-6 Astra halved research time and code cost; invideo reports about 3x color-grading success.

OpenAI published two customer accounts of GPT-6 Astra on consecutive days. On 2026-09-22 it said Parallel’s AI agents reached high-quality research in half the time and cut code cost by 50%, using fewer steps and more targeted searches. On 2026-09-23, invideo said Astra plans complex video edits at frame-level accuracy with fewer reasoning steps and output tokens than earlier models it tested. CEO Sanket Shah reported color-grading and color-correction success about three times higher, and a few editors built roughly 50 custom, fully editable effects in one day. The sources describe different workloads and do not contradict each other; both are company-reported results rather than independent benchmarks.

• On 2026-09-22, OpenAI said customer Parallel used GPT-6 Astra so AI agents produced high-quality research in half the time with a 50% code-cost reduction.
• Parallel reported fewer steps, more targeted searches, and better agent delegation on demanding research tasks, at the same stated quality.
• On 2026-09-23, agentic video editor invideo said GPT-6 Astra plans complex edits at frame-level accuracy with fewer reasoning steps and output tokens than models it previously tested.
• invideo CEO Sanket Shah said color-grading and color-correction success rates improved about threefold.
• A few invideo editors used Astra to build roughly 50 custom, fully editable effects in a single day.
• The two reports cover different products and workloads and do not state conflicting figures.