Agentic data science with the Data Agent Kit
cloud.google.com·6h ago
Data Science
Simon Willison's Blog·Sep 3·5 min read
GPT‑6 Astra
TL;DR
GPT-6 Astra is being rolled out to select organizations and will soon be available to all ChatGPT Plus, Pro, Business, and Enterprise users. It outperforms competitors on benchmarks, achieving a remarkable 99.9% on the ARC-AGI 3 benchmark.
✦ Why It Matters
Evaluate how GPT-6 Astra can enhance your organization's AI capabilities, especially in security tasks.
Key Takeaways
- Explore GPT-6 Astra for advanced AI applications in your projects.
- Consider the pricing model for budget planning on API usage.
- Leverage Astra's security capabilities for enhanced protection.
- Stay updated on performance benchmarks to inform your AI strategy.
Full Summary
GPT-6 Astra is OpenAI's latest model, designed to compete with Claude Fable 5 and 5.1, and is set to be available through various subscription tiers and APIs. It is priced at $10 per million input tokens and $50 per million output tokens.
Astra has achieved an impressive 99.9% score on the ARC-AGI 3 benchmark, although this was accomplished using a custom 'Provider Adapter harness' that optimizes performance by maintaining context across requests. In security tasks, Astra excels, scoring 100% on ExploitBench and significantly outperforming previous models in binary reverse engineering tasks.
These results suggest that Astra could be a powerful tool for organizations needing advanced AI capabilities, particularly in security-related applications.