Neura Intelligence Index
Neura Intelligence Index
AI Model Benchmarks
Compare AI model performance across 48+ benchmarks. Our composite index aggregates coding, math, reasoning, and language scores into a single intelligence ranking.
372
Models Ranked
48
Benchmarks
Jul 21, 2026
Last Updated
Highlights
At-a-glance rankings across the three dimensions that matter most.
Intelligence
Neura Intelligence Index; Higher is better
Context Window
Max input tokens; Higher is better
Price
USD per 1M output tokens; Lower is better
Leaderboard
Ranked by the Neura Intelligence Index — a weighted composite of 48 benchmarks across 8 categories.
Intelligence Tradeoffs
Find the sweet spot — models in the top-left quadrant offer the best value.
Intelligence vs. Output Cost ($/1M tokens)
Google
DeepSeek
Zhipu AI
StepFun
Alibaba Cloud / Qwen Team
OpenAI
Amazon
Moonshot AI
Anthropic
Meta
xAI
MiniMax
ByteDance
Mistral
Inception
Xiaomi
NVIDIA
Meituan
Microsoft
Mistral AI
Alibaba
Cohere
Category Leaders
Top performers in each benchmark category.
General Knowledge
Coding
Reasoning
Multimodal
Top Models per Task
Who leads on each individual benchmark — click any card to see full results.
Best Model For Your Use Case
Different tasks need different strengths. These indices re-weight our benchmarks for specific workflows.
Coding & Engineering
Best for software development, code generation, and debugging
Research & Analysis
Best for scientific research, data analysis, and complex reasoning
Content & Copywriting
Best for writing, editing, summarization, and creative tasks
Best Value
Highest intelligence per dollar — the cost-efficiency sweet spot
Methodology
The Neura Intelligence Index is a composite score (0-100) computed from 48+ individual benchmarks spanning 8 categories: coding, math, reasoning, general knowledge, language, multimodal, safety, and agentic tasks.
How it works
- Normalize — Each benchmark score is min-max normalized to 0-1 across all models.
- Categorize — Normalized scores are averaged within their benchmark category.
- Weight — Category averages are combined using weights: Coding (20%), Reasoning (20%), General (15%), Math (15%), Language (10%), Multimodal (10%), Safety (5%), Agent (5%).
- Scale — The weighted average is scaled to 0-100.
Coverage & confidence
Not all models have scores on all benchmarks. The confidence indicator reflects benchmark coverage: high (>70% of benchmarks), medium (40-70%), or low (<40%). Weights are renormalized across available categories so models aren't penalized for missing data.
Data sources
Scores are aggregated from official model cards, Papers With Code, HuggingFace Open LLM Leaderboard, LiveBench, and LMSYS Chatbot Arena. Each score includes a verification status (official, self-reported, or aggregated).
Frequently Asked Questions
What is the Neura Intelligence Index?▾
The Neura Intelligence Index is a composite score (0-100) that aggregates AI model performance across 15+ benchmarks spanning coding, math, reasoning, general knowledge, language, multimodal, safety, and agentic tasks. It uses min-max normalization and weighted category averaging to produce a single comparable score.
How often are benchmark scores updated?▾
Benchmark scores are synced daily via an automated pipeline that aggregates data from official model cards, Papers With Code, HuggingFace Open LLM Leaderboard, LiveBench, and LMSYS Chatbot Arena.
What does the confidence level mean?▾
Confidence reflects how many benchmarks a model has been tested on relative to the total. High means >70% benchmark coverage, medium is 40-70%, and low is <40%. Models with low coverage may have composite scores that shift as more benchmarks are added.
How are category weights determined?▾
Category weights reflect real-world demand: Coding and Reasoning each get 20%, General and Math each get 15%, Language and Multimodal each get 10%, and Safety and Agent each get 5%. Weights are renormalized across available categories so models are not penalized for missing data.