OpenAI Pitches GPT-6 Astra for Business, Claiming 4x Human Speed in the Excel World Cup

OpenAI published a business-focused companion post to GPT-6 Astra’s launch, framing the model around dropping into existing enterprise workflows “from day one, without extensive preparation or engineering work.”

Details

• Works with what’s already there: Astra can operate through standard business applications that lack APIs — automating tasks across Excel, analytics platforms, and other desktop tools without custom integration work
• Professional task performance: strong results on financial modeling, code generation, data analysis, and document production, while sticking to corporate branding standards and templates
• New enterprise plugins: integrations with Oracle Analytics, Power BI, Navan, and Avalara
• Standout benchmark: completes Excel World Cup challenges about four times faster than the competition’s winning human contestant
• Cost and coding performance: priced at $10/$50 per million tokens, scoring 57.9% on Terminal-Bench 4.0 versus 37.3% for the prior model, at a lower effective API cost
• Safety in practice: produces unintended outcomes 89% less often than GPT-5.6 Sol on OpenAI’s internal computer-use safety benchmarks
• Internal validation: OpenAI says its own engineering teams achieved 25x lower turn latency by using Astra to optimize their workflows

What happened next

The post doubles down on administrative controls, approval workflows, and cybersecurity alignment as the differentiators meant to win over enterprise buyers, positioning Astra’s business pitch around low-friction adoption rather than requiring companies to rebuild processes around it.