SAI Framework — Sustainable AI Intelligence Framework

Version: 0.2.1 (Draft)
Date: July 2026
License: CC-BY 4.0

Overview

The Sustainable AI Intelligence (SAI) Framework is an open specification for evaluating Large Language Models (LLMs) across three dimensions — Token Efficiency, Intelligence/Value Generation, and Energy Sustainability — and combining them into the Sustainable Intelligence Index (SII).

This repository contains the normative and informative documents that define the standard. It is not an application, SDK, or provider integration.

Core Dimensions

Dimension	Name	Primary Metric
X	Token Efficiency	Normalized tokens per intelligence unit (X_norm)
Y	Intelligence / Value	Weighted benchmark score (0–100)
Z	Energy Sustainability	Intelligence Points per Joule (IP/J)

SII: SII = (Y × Z) / X_norm × 100

Full definitions are in SPEC.md.

Specification Documents

Document	Role
SPEC.md	Normative core — metrics, SII, governance
BENCHMARK_GUIDE.md	Normative — benchmark execution and Y scoring
MEASUREMENT_PROTOCOL.md	Normative — energy measurement (Z dimension)
CONFORMANCE.md	Normative — SAI-Basic / SAI-Full requirements
schemas/evaluation_report.yaml	Report schema for published evaluations

Informative Examples

Document	Role
examples/worked_example.md	End-to-end evaluation walkthrough
examples/sii_calculator.py	Reference implementation (Python stdlib only)

python3 examples/sii_calculator.py
python3 examples/sii_calculator.py --self-test

Benchmark harnesses, inference APIs, and hardware tooling belong in separate projects that conform to this specification.

Workload Categories

• General / Reasoning — business analysis, writing, QA (default)
• Coding — software development, debugging
• Scientific / Mathematical — research, calculations
• Agentic / Tool-Use — autonomous agents, automation

Conformance Levels

• SAI-Basic v0.2.1 — ≥70% benchmark coverage, Tier 2+ energy, suite-scoped Z
• SAI-Full v0.2.1 — 100% primary coverage, Tier 1 energy, ≥2 workload categories
• SAI-Certified — planned for v1.0 (independent verification)

See CONFORMANCE.md.

Web3 Tokenomics (Informative)

SAI defines a bridge between AI token consumption and verifiable intelligence units. Detailed Web3 integration is planned for v0.3.

Governance and Contributions

See CONTRIBUTING.md and ROADMAP.md.

Related Work

• TokenPowerBench
• Hugging Face Open LLM Leaderboard
• ML.Energy Leaderboard
• MLPerf, TPCx-AI

License

CC-BY 4.0