AIPI (ATOM Inference Price Index) methodology was developed by information economists with 25+ years advising enterprise C-suite teams on pricing intelligence and revenue strategy. AIPI follows financial index methodology standards comparable to S&P, MSCI, and Bloomberg indexes.
Why Investment-Grade
AIPI meets institutional investment standards through rigorous methodology that ensures transparency, reproducibility, and accuracy:
- Deterministic extraction — Rules-based logic, not AI estimation or web scraping
- Reproducible calculations — Same inputs always produce same outputs
- Composition-adjusted — Chained matched-model methodology isolates actual price movements from vendor model mix changes
- Transparent methodology — Publicly documented construction process
- Audit trail — All data points tracked from source to index
- Human verification overlay — Analyst review for edge cases and validation
Data Quality Standards
Every price in AIPI undergoes rigorous quality controls to ensure institutional-grade accuracy:
| Standard |
Implementation |
| Deterministic Extraction |
Every price is extracted using rules-based logic programmed to recognize vendor-specific pricing page formats. No AI interpretation, no estimation, no guesswork. |
| Zero Estimation |
We never interpolate, extrapolate, or estimate pricing. If a vendor does not publish a price, it is not included in the index. |
| Human Verification |
Information economists manually review: (1) New vendor pricing page formats, (2) Ambiguous pricing structures, (3) Price changes exceeding 20% week-over-week, (4) Regional availability verification |
| Audit Trail |
Original and normalized prices are stored with extraction timestamps, enabling full verification of any index value back to source data. |
| 24-Hour Freshness |
Automated monitoring detects vendor pricing updates and processes changes within 24 hours of publication. |
Historical Data Depth
AIPI historical data is available from December 2024 forward. For vendors onboarded after this date, pricing history begins at the vendor's onboarding date. Historical data is updated retroactively when vendors publish prior period pricing information.
| Period |
Coverage |
| December 2024 - Present |
Full coverage for all vendors onboarded by December 2024 |
| New Vendors |
History begins at onboarding date; retroactive backfill if vendor provides historical pricing |
| Index Calculation |
Indexes are calculated for all periods with sufficient vendor representation (minimum 3 vendors) |
Industry Comparison
AIPI provides a fundamentally different approach to AI pricing intelligence compared to traditional alternatives:
| Source |
Update Frequency |
Methodology |
Coverage |
| Industry Reports |
Monthly or Quarterly |
Analyst research, surveys, selective vendor sampling |
Limited (5-15 vendors) |
| Analyst Surveys |
Quarterly |
Self-reported data, voluntary participation |
Variable |
| Vendor Websites |
Ad-hoc |
Manual checking, no normalization |
Single vendor only |
| AIPI Public |
Weekly (Automated) |
Deterministic extraction, normalized, indexed |
14 benchmark indexes |
| AIPI Premium |
Daily (Automated) |
Deterministic extraction, normalized, indexed |
-- SKUs, -- vendors |
Coverage Growth
ATOM continuously expands coverage as new vendors enter the AI inference market and existing vendors expand their offerings:
| Metric |
Current Status |
| SKU Coverage |
-- inference SKUs tracked globally |
| Vendor Coverage |
-- AI vendors across model developers, inference platforms, cloud marketplaces, and neocloud providers |
| Vendor Classification |
4 channel types: Model Developers (direct API), Cloud Marketplaces (Bedrock, Vertex, Azure), Inference Platforms (DeepInfra, Fireworks, Together AI), Neoclouds (Groq, Cerebras) |
| Onboarding Rate |
2-4 new vendors per month, prioritizing market leaders and high-growth platforms |
| Expansion Roadmap |
Targeting 50+ vendors and 2,000+ SKUs by Q2 2026 |
The AIPI (ATOM Inference Price Index) tracks AI inference costs across
-- vendors globally, providing developers, analysts, and infrastructure buyers with transparent pricing intelligence through 14 benchmark indexes.
Index Categories
| Category |
What It Tracks |
Key Question |
| Modality |
AI capability type (text, image, audio, video, voice, multimodal) |
What does this type of inference cost? |
| Channel |
Distribution channel in the supply chain |
Where should I buy inference, and where do arbitrage opportunities exist? |
| Tier |
Model capability level |
What is the premium for flagship intelligence versus budget alternatives? |
| Licensing |
Open-source vs proprietary models |
How much cheaper is open-source inference across all channels? |
Modality Indexes
All modality indexes are calculated globally across all tracked vendors, regardless of vendor origin or region.
| Index Code |
Description |
Unit |
| AIPI TXT GLB |
Text generation models from leading AI vendors globally |
per 1,000 tokens |
| AIPI MML GLB |
Multimodal models from leading AI vendors globally |
per 1,000 tokens |
| AIPI IMG GLB |
Image generation models from leading AI vendors globally |
per image |
| AIPI AUD GLB |
Audio transcription and generation models from leading AI vendors globally |
per minute |
| AIPI VID GLB |
Video generation models from leading AI vendors globally |
per second |
| AIPI VOC GLB |
Voice and speech synthesis models from leading AI vendors globally |
per 1,000 characters |
Channel Indexes
Channel indexes track text model pricing (per 1,000 tokens) across four distinct distribution channels globally. These indexes enable direct comparison of the same or comparable models across different points of sale, surfacing arbitrage opportunities and channel-specific pricing dynamics.
| Index Code |
Description |
Channel |
| AIPI DEV GLB |
Text models priced directly by model developers globally |
Direct API (OpenAI, Anthropic, Google, Mistral, etc.) |
| AIPI CLD GLB |
Text models priced through cloud marketplaces globally |
Cloud Marketplace (AWS Bedrock, Google Vertex, Azure) |
| AIPI PLT GLB |
Text models priced through third-party inference platforms globally |
Inference Platform (DeepInfra, Fireworks, Together AI) |
| AIPI NCL GLB |
Text models priced through neocloud providers globally |
Neocloud (Groq, Cerebras) |
Tier Indexes
Tier indexes track text model pricing (per 1,000 tokens) segmented by model capability level globally.
| Index Code |
Description |
Scope |
| AIPI FTR GLB |
Top-tier flagship models from leading AI vendors globally |
Flagship releases |
| AIPI BDG GLB |
Low-cost economy models from leading AI vendors globally |
Budget variants |
| AIPI RSN GLB |
Chain-of-thought reasoning models from leading AI vendors globally |
Reasoning models |
Licensing Index
The licensing index tracks the pricing of open-source and open-weight models across all distribution channels globally, measuring the cost advantage of open-source inference relative to the broader market.
| Index Code |
Description |
Unit |
| AIPI OSS GLB |
Open-source and open-weight models across all channels globally |
per 1,000 tokens |
Data Sources
| Method |
Description |
| Direct Extraction |
Automated extraction from official vendor pricing pages using deterministic rules-based logic |
| API Ingestion |
Direct ingestion via vendor APIs where available, ensuring real-time accuracy |
| Analyst Curation |
Information economists manually review new vendor pricing formats, ambiguous pricing structures, price changes exceeding 20%, and regional availability verification |
The AIPI (ATOM Inference Price Index) is built through a structured pipeline:
qualifying data, excluding non-comparable pricing, normalizing units and currencies, and calculating
composition-adjusted index values.
Inclusion Criteria
To be included in AIPI, a SKU must meet all of the following:
| Criterion |
Requirement |
| Publicly listed |
Price must be published on the vendor's website |
| Pay as you go |
Standard on demand pricing only |
| Production ready |
Generally available models only |
| Matching unit |
Price must be convertible to the index's normalized unit |
Exclusions
The following pricing types are excluded from AIPI:
| Pricing Type |
Reason |
| Negotiated rates |
Not publicly verifiable |
| Enterprise contracts |
Not publicly verifiable |
| Committed use discounts |
Not comparable to on demand pricing |
| Volume tiers |
Only base tier (lowest volume) is indexed |
| Batch pricing |
Not comparable to real-time inference pricing |
| Subscriptions |
Not comparable to usage based pricing |
| Bundled models |
Cannot attribute price to single modality |
| Free tiers and trials |
Not representative of production costs |
| Beta and preview models |
Pricing may change at general availability |
| Legacy models |
No longer actively offered |
Normalization
Vendors publish prices in different formats, units, and currencies. AIPI normalizes all prices to enable direct
comparison.
| Vendor Format |
Normalized To |
Conversion |
| per 1M tokens |
USD per 1K tokens |
÷ 1,000 |
| per 1K tokens |
USD per 1K tokens |
No change |
| per image |
USD per image |
No change |
| per million pixels |
USD per image |
× (resolution ÷ 1M) at 1080p |
| per second (video) |
USD per second |
No change |
| per million pixels (video) |
USD per second |
× (resolution ÷ 1M) × fps at 720p 24fps |
| per minute (audio) |
USD per minute |
No change |
| per second (audio) |
USD per minute |
× 60 |
| per 1K characters |
USD per 1K characters |
No change |
| per character |
USD per 1K characters |
× 1,000 |
| per page |
USD per page |
No change |
Currency conversion: For CNY pricing, ATOM uses a fixed annual exchange rate (set January 1st, held for 12 months) rather than daily spot rates. This eliminates foreign exchange volatility from price trend analysis, isolating genuine AI pricing movements from currency fluctuations. Chinese vendors operate in a relatively stable pricing environment, and removing FX noise enables clearer analysis of strategic pricing decisions. All other currencies use daily spot rates as their pricing stability is well-established. The database stores both original and normalized prices for audit purposes.
Calculation Method: Chained Matched-Model
AIPI indexes use a chained matched-model methodology to isolate actual vendor price movements from changes in the composition of tracked models. This approach is standard practice in financial index construction (comparable to CPI matched-model methodology) and prevents model additions or removals from creating false price signals.
The problem it solves: In a rapidly evolving market, vendors frequently add new models and retire old ones. A simple average would shift whenever the model mix changes, even if no vendor actually changed a price. For example, if three expensive models are delisted in a given week, a simple average would show a price decline that never occurred. The matched-model approach eliminates this composition bias entirely.
How it works:
| Step |
Description |
| 1. Base week |
The earliest week in the dataset establishes the index level using a simple unweighted average of all qualifying SKUs. |
| 2. Match |
For each subsequent week, identify the matched set: SKUs present in both the current and prior week. New additions and removals are excluded from the comparison. |
| 3. Measure |
Compute the percentage change in the average price of the matched set between the two weeks. This captures only genuine vendor repricing. |
| 4. Chain |
Apply that percentage change to the prior week's index value. The index level evolves continuously through chained multiplication. |
| 5. Absorb |
New models enter the matched set the following week. Removed models exit silently. Neither event affects the index level at the point of change. |
Base Week: Index(t=0) = AVERAGE(Price) for all qualifying SKUs
Subsequent: Matched_Change = AVG(Price_current) / AVG(Price_prior) for matched SKUs only
Index(t) = Index(t-1) × Matched_Change
Why unweighted: Within each matched set, all vendors and models are weighted equally. This prevents market dominance by high-volume providers from skewing the index and ensures pricing trends from emerging vendors are visible alongside established players. The unweighted approach provides a balanced representation of the full pricing spectrum available to buyers.
Each index is calculated separately for three pricing directions:
| Direction |
What It Measures |
| Input |
Cost to send tokens or data to the model |
| Cached Input |
Discounted rate for repeated or cached context |
| Output |
Cost for tokens or data generated by the model |
AIPI Volatility Composite
The AIPI Volatility Composite measures week-over-week pricing stability across AI inference markets. Unlike the price indexes that track absolute levels, this composite tracks the percentage of models that changed price from one week to the next.
Calculation methodology: For each pricing direction (Input, Cached Input, Output), the system identifies models present in both the current and prior week (the matched set), then counts how many of those models changed price. The volatility percentage is calculated as:
Volatility % = (Matched Models with Price Changes ÷ Total Matched Models) × 100
Key features:
| Feature |
Description |
| Matched-model basis |
Only models present in consecutive weeks are compared, ensuring new additions and removals do not inflate volatility |
| Model-level tracking |
Tracks changes at the model level to avoid triple-counting a single model repricing across Input, Cached, and Output |
| Directional splits |
Three separate volatility measures (Input, Cached Input, Output) reveal which pricing components are most dynamic |
| Zone classification |
Volatility levels are classified into zones: 0-2% STABLE, 2-4% HIGH, 4-6% EXTREME |
Investment implications: Volatility levels above 2% historically coincide with major market events such as new flagship model launches, competitive repricing waves, or capacity constraints. The composite serves as an early warning system for pricing instability that may impact infrastructure budgets, vendor negotiations, or portfolio company cost structures. Institutional buyers use volatility trends to time vendor contract negotiations and assess competitive dynamics.
Benchmark context: Enterprise B2B API providers (Stripe, AWS, Twilio) typically maintain 99.5%+ weekly stability (under 0.5% volatility). AI inference markets show structurally higher volatility due to rapid innovation, competitive dynamics, and capacity optimization. The AIPI Volatility Composite provides a quantitative measure of this market characteristic, enabling investors and operators to monitor pricing stability trends over time.
Update Schedule
| Activity |
Public (free) |
Premium (paid) |
| Price extraction |
Weekly (Monday) |
Daily |
| Index calculation |
Weekly (Monday) |
Daily |
| Dashboard refresh |
Instant |
Instant |
| Price change alerts |
Not available |
Instant |
| New vendor onboarding |
As available |
As available |
| Tier index review |
Quarterly |
Quarterly |
Data freshness: Every SKU carries a Last Verified Date, the date
when pricing was last confirmed against the vendor's published page. When a vendor updates pricing, ATOM's
automated pipelines detect the change and update the database within 24 hours.
The AIPI (ATOM Inference Price Index) uses standardized terminology to ensure
consistency across all indexes, reports, and data exports.
Glossary
| Term |
Definition |
| SKU |
Stock Keeping Unit, a unique price point for a specific model, vendor, and direction combination |
| Index |
A benchmark measure calculated from composition-adjusted average prices of all SKUs matching specific criteria |
| Matched Set |
The group of SKUs present in both the current and prior week, used to calculate week-over-week price changes without composition bias |
| Chained Index |
An index value derived by applying matched-set percentage changes to the prior period's level, isolating genuine price movements from model mix changes |
| Normalized Price |
A vendor's price converted to AIPI's standard unit and currency for comparability |
| Direction |
Whether a price applies to Input tokens, Cached Input tokens, or Output tokens |
| Token |
A unit of text processing, roughly 4 characters or 0.75 words in English |
| Input Tokens |
Tokens sent to the model (prompt, context, instructions) |
| Output Tokens |
Tokens generated by the model (response, completion) |
| Cached Input |
Discounted rate when input tokens are reused from a previous request |
| Modality |
The type of AI capability (Text, Image, Audio, Video, Voice, Multimodal) |
| Channel |
The distribution pathway through which inference is sold: Model Developer (direct API), Cloud Marketplace (Bedrock, Vertex, Azure), Inference Platform (DeepInfra, Fireworks, Together AI), or Neocloud (Groq, Cerebras) |
| Neocloud |
AI-specialized cloud provider offering serverless per-token inference on custom or purpose-built silicon, distinct from traditional GPU-based inference platforms |
| License Type |
Classification of model weight availability: open-source (weights publicly available for download and self-hosting) or proprietary (API-only access) |
| Tier |
Model capability classification: Frontier (flagship), Budget (economy), or Reasoning (chain-of-thought) |
| Inference |
Running a trained AI model to generate outputs |
| Last Verified Date |
The date when a SKU's pricing was last confirmed against the vendor's published page |
| Volatility |
The percentage of matched models that changed price from one week to the next |
Scope and Focus
AIPI is deliberately optimized for investment-grade pricing intelligence. The following capabilities define what AIPI is designed to deliver:
| AIPI Optimizes For |
Description |
| Price Transparency |
Clear, normalized pricing data enabling direct vendor comparison |
| Market-Wide Trend Detection |
Identifying pricing movements, competitive dynamics, and market shifts |
| Vendor Positioning Analysis |
Understanding competitive positioning through pricing strategy |
| Budget Forecasting Accuracy |
Reliable data for AI infrastructure cost planning and modeling |
AIPI indexes measure price only. The following dimensions are explicitly not covered, as they require different methodologies and data sources:
| Out of Scope |
Why |
| Performance or latency |
AIPI tracks cost, not speed or throughput; performance benchmarks require separate infrastructure |
| Quality benchmark scores |
AIPI tracks model specifications (parameters, context window, capabilities, tool support) but does not score output quality; evaluation benchmarks like MMLU or HumanEval require separate testing frameworks |
| Total cost of ownership |
Excludes infrastructure, egress, support, and integration costs which vary by deployment architecture |
| Fine tuning or training |
Only inference (running models) is tracked; training costs follow different pricing structures |
