NotebookLM
Model
Cohort
May '25
Cohort
June '25
Retention
retention
retention
A Cautionary Tale: No Fit, No Retention
The "Boomerang Effect": Some Churned Users Return
e.g.Gemini 2.5 ProJune '25 e.g., Claude 4 Sonnet (May'25
FOUNDATIONAL COHORTS
Llama 4 Maverick
DeepSeck
Models that that solve a critical workload create a "foundational cohort" with high retention.
The "Glass Slipper" Effect Explains User Stickiness
Who Stays, Who Leaves: The "Cinderella Glass Slipper" Effect
(Low Cost, Low Usage)
LONG TAIL
GPT-3Pro
OpenAl
(High Cost, Low Usage)
PREMIUM SPECIALISTS
Google Gemini Flash
EFFICIENT GIANTS (Low Cost, High Usage)
Anthropic Claude Sonnet 4
PREMIUM LEADERS (High Cost,High Usage)
Different Models, Different Jobs
R
eepestos')
(Volume)
USAGE
Technology,Finance, Health, have the highest cost-per-token, as users will pay more for high accuracy in critical domains.
High-Stakes Workloads Command Premium Prices
High
Cost
Low
Low
Prompts Are Getting 4x Longer, Driven by Coding
High
Superior capabilities often matter more than price alone
Demand is Surprisingly Price-Inalastic
Tools
Web
The Economics of Al: Cost, Usage & Market Segments
<158
OpenAl
Qwen
PROGRAMMING
Traditional text generation
50%
OpenAl
Meta
(MEDIUM)
50%
REASONING MODELS
15B-70B
Meta
Qwen
~60% spend capture
Approximately 15% of interactions invake external tools
Al
Tool Use is Becoming Mainstream
Reasoning Models Now Power Over 50% of All Usage
Usage sweet spot for capability & efficiency
The OsS Market Is Fragmenting
The Rise of the "Medium" Model
OPEN-SOURCE (OSS) MODELS
Multi-step, tool-integrated workflows where LLMs plan, reason, and act to accomplish complex goals.
CHINESE OSS MODELS
ROLEPLAY &CREATIVE
Anthropic's Claude Dominates the Coding Arena
What is Agentic Inference?
Over 52% of all open-source model usage
The New Paradigm: Rise of Agentic Inference
Chinese OSS Models Are a Major Growth Engine
Surprise Winner: Roleplay is the Killer App for Open Source
(over S0% of total token volume)
Programming Is Now the #1 Use Case Overall
What 100 Trillion Tokens Reveal About Real-World LLM Usage
30%
Steady growth to 30% by late 2025
A Dual Ecosystem: OSS Models Capture One-Third of the Market
What Are People Using LLMs For?
The State of Al
The Shifting Landscape: Open vs. Closed Models