An open-weight model, Step 5 Preview, demonstrates a significant performance gap of 28.4 points compared to the proprietary GPT-6 Astra. Despite this difference, Step 5 Preview is remarkably more cost-effective, costing 19 times less per million output tokens. This cost advantage challenges the traditional understanding of what it means for a model to be considered "behind" in terms of capability versus accessibility. AI
IMPACT Highlights the trade-offs between performance and cost in AI models, suggesting open-weight models may offer compelling value propositions.
RANK_REASON The item discusses a comparison between two models but does not originate from a primary source announcement of either model.
Read on Mastodon — mastodon.social →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →