OpenAI Lays Out Its 'Abundant Intelligence' Strategy: Cheaper GPT-5.6 Pricing and a Four-Layer Infrastructure Approach

OpenAI has published a company-wide strategy piece called “Building Abundant Intelligence,” laying out its approach to making AI more capable, more affordable, and more widely available. The post explains how its infrastructure, models, platform, and product layers reinforce one another, and discloses concrete GPT-5.6 price cuts along with recent efficiency gains.

Details

• Core philosophy: OpenAI frames the value of infrastructure investment not in terms of scale but of outcomes — “more capable intelligence, available to more people, at a lower cost.” Better intelligence drives adoption, adoption generates revenue and feedback, and reinvesting that feedback further improves capability and efficiency
• Price cuts: GPT-5.6 Luna gets an 80% price cut, to $0.20 per million input tokens and $1.20 per million output tokens. GPT-5.6 Terra drops 20%, to $2 and $12 respectively. GPT-5.6 Sol gains a “Fast mode” that runs 2.5x faster at twice the standard price, with no reduction in intelligence
• Efficiency gains: Production software optimizations cut end-to-end serving costs by 20%, and improved speculative decoding raised token-generation efficiency by more than 15%. On ARC-AGI-3, scores rose from 13.3% to 38.3% — using six times fewer output tokens — purely through better reasoning retention and context management, with no change to the underlying model
• Scale: ChatGPT now serves more than one billion active users and over two million businesses. Daily messages per user rise roughly 50% within six months of signup, and the range of work categories people use ChatGPT for roughly doubles over that period
• Enterprise and agentic use: ChatGPT Work is shifting knowledge work “from asking to doing,” and agentic work in Codex now accounts for 99.8% of its weekly output tokens. Teams like finance are folding agentic tools into core workflows
• Four-layer approach: OpenAI describes operating across four integrated layers — infrastructure, models, platform, and products — where “each layer makes the others better.” Feedback from real products shapes research priorities, and research improvements in turn strengthen products and lower serving costs
• Investment discipline: Capacity decisions are driven by user and workload growth, enterprise commitments, API consumption, utilization, revenue, and progress in model capability and efficiency. The stated goal is to “deploy the right capacity, at the right time, against credible demand,” rather than maximizing infrastructure for its own sake
• Build-vs-buy: OpenAI says it doesn’t need to own every layer itself, choosing to “own, partner, or buy depending on what best serves the customer” while coordinating the system as a whole
• Future outlook: OpenAI expects more capable systems to complete longer projects, coordinate across tools, and take on more end-to-end work — extending capabilities once reserved for large enterprises to individuals and smaller organizations

How to try it

• Check the new GPT-5.6 pricing and modes (including Sol’s Fast mode) in ChatGPT, ChatGPT Work, and Codex
• If cost efficiency matters most, look at the reduced Terra or Luna pricing; if speed matters most, consider Sol’s Fast mode
• Read OpenAI’s full strategy post at https://openai.com/index/building-abundant-intelligence