Observed Signal · Apr 26, 2026 · Technical Release · Source: Exponential View · Impact: 4/5 · Sentiment: Positive
Intelligence Per Token: The New AI Metric
The newsletter argues that as inference compute becomes a binding constraint, the industry should compare AI models by intelligence delivered per token or per dollar rather than by a single benchmark score. The author cites a tweet from OpenAI reasoning lead Noam Brown after GPT-5.5’s rollout, and contrasts US labs’ 'more compute' culture with Chinese labs that optimize for compute scarcity. DeepSeek’s V4 model is highlighted as marginally lower-performing than GPT-5.4 but roughly 4x cheaper, illustrating a shift toward inference-efficiency. The piece notes inference costs are rising in importance (approaching ~10% of engineering headcount spend) and that compute economics will shape model design, deployment and competitiveness.
Major-model rollout (GPT-5.5) and the industry shift toward inference-efficiency materially affect model selection, deployment economics, and infrastructure decisions across AI-dependent tech sectors.
Track DeepSeek Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your Polaris7 workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- OpenAI rolled out GPT-5.5 (mentioned in the newsletter).
- Noam Brown (identified as OpenAI’s reasoning research lead) tweeted that intelligence is a function of inference compute and that models should be compared by intelligence per token or per dollar.
- DeepSeek’s V4 model is described as marginally worse than GPT-5.4 but approximately 4x cheaper.
- The newsletter states inference costs are approaching about 10% of total engineering headcount spend, making inference economics material to teams and product decisions.
- Chinese AI labs are adapting model design to compute scarcity, treating compute constraints as strategic design specifications.
Connected Companies & Entities
Ontology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Inference Engineering: The New Tokenomics of AI
The article, a paid newsletter piece, argues that the AI industry is shifting from a training-centric to an inference-centric phase. It explains that as models become more capable, the economic focus moves to the continuous operation of AI across enterprise workflows. The piece details the technical and economic distinctions between prefill (reading) and decode (writing) stages of inference, and introduces concepts like KV cache management, batching, prefix caching, and latency considerations. The author predicts that enterprise inference will become economically as important as pretraining, and that the optimization goal changes from lowest token cost to lowest cost per accepted outcome at required latency. The article is primarily an analytical commentary, not a news report, and is likely paywalled as indicated by 'Subscribe to Premium to Gain Access'.
The Weird Economics of AI Tokens
The article analyzes how modern AI billing and infrastructure have made “tokens” the primary economic unit of intelligence. It explains that tokens (pieces of text processed by models) are a useful billing abstraction but differ widely in computational cost and economic value: input vs output tokens, short vs reasoning-heavy requests, and token usage vs usefulness. The piece describes data centers as “token factories,” highlights risks from subscription and context-window costs, and argues that model routing, token observability, and measuring cost per useful task (not cost per token) will shape AI economics going forward.
Measuring AI Value: Useful Intelligence per Dollar
OpenAI outlines a framework for assessing AI economics centered on a proposed metric, "Useful Intelligence per Dollar," which asks whether AI completes valuable work, how much successful tasks cost, how dependable results are, and whether value improves at scale. The piece argues businesses should measure end-to-end cost per successful task (including retries, human review, and employee time) rather than cost per token, and track dependability categories (ready to use, needs correction, needs escalation). OpenAI also describes infrastructure and compute as central to improving model capability and efficiency. The post announces GPT‑5.6 (three tiers: Sol, Terra, Luna), claims GPT‑5.6 Sol set a new state of the art on certain long-horizon engineering benchmarks while using fewer output tokens, and positions ChatGPT Work and ChatGPT Enterprise as enterprise offerings built on OpenAI's security, privacy, and compliance foundations.
Receive all future signals for DeepSeek & competitors in your briefing
Never miss material M&A deals, leadership changes, account wins, or product rollouts. Polaris7 monitors 20,000+ primary sources 24/7, delivering verified telemetry directly into your executive briefing.
