875 models • 711 with benchmarks • 664 with AA Score • Synced: October 16, 2026
What is the best LLM today?
Most Intelligent
AA Score — Artificial Analysis · updated every 6h
Fastest?
Fastest
Output tokens/second · higher is better
Best value for money?
Most Value per Dollar
AA Score per US$ (blended 3:1)
🏆 Intelligence Ranking
Score AA — Artificial Analysis · top 20
AnthropicAnthropicAnthropic
MetaAnthropicAnthropicAnthropic
OpenAI
GoogleMetaScore AA = AA Intelligence Index from Artificial Analysis. Updated every 6h. Click any model for detailed benchmarks.
875
Models
107
Companies
664
With AA Score
99
Reasoning
126
Open Source
194
Multimodal
Ranking by AA Score
Classification based on the AA Intelligence Index by Artificial Analysis — composite score updated every 6 hours.
What is an AI benchmark?
An AI benchmark is an independent, standardized test that measures and compares the real-world capability of artificial intelligence models across specific tasks. Unlike marketing claims from AI companies, benchmarks provide objective, verifiable metrics that help developers and businesses choose the right model for their use case.
The 3 criteria that matter most when comparing AI models:
- Intelligence (AA Score): Composite index aggregating GPQA Diamond, MMLU-Pro, AIME, HLE and LiveCodeBench — measures reasoning, math, science and coding capability.
- Speed (tokens/second): How fast the model generates text — critical for real-time applications like chatbots and coding assistants.
- Cost (USD/1M tokens): Price per million tokens — determines whether a model is viable for production at scale.
🔰 What do ELO, Score AA and tokens/s mean? Quick guide
ELO: Human preference ranking from LMArena (Chatbot Arena). Real users compare blind model responses and vote. Higher = people prefer it for everyday conversations.
Intel. (AA Intelligence Index): Composite score 0-100 by Artificial Analysis aggregating GPQA Diamond, MMLU-Pro, AIME, HLE, LiveCodeBench and more. Measures real technical capability — reasoning, math, coding, science.
Input $/1M: Price in USD per 1 million input tokens (≈750,000 words). What you pay to send text to the model. Output tokens cost separately.
Context: Maximum tokens the model can process at once — determines how much text (code, documents, conversation history) fits in a single request.
Tokens/s: Generation speed — how many tokens the model outputs per second. Higher = faster responses, lower latency, better for real-time apps.
Prices in USD per 1M input tokens. See methodology for details.
OS = Open Source • MM = Multimodal • R = Reasoning •AA Score: Artificial Analysis • Intel.: Artificial Analysis •Pricing: OpenRouter •View full methodology
Intelligence over time
Daily progression of the AA Intelligence Index for top models. Higher = more capable.
⚡ Inference Speed
Tokens per second — top 15
GoogleGoogleMetaGoogleSpeed measured in tokens/second via API. TTFT = Time to First Token (latency until the first response).
Models by Company
AI Benchmark Guide 2026
Claude Opus 5.5 (Adaptive Reasoning, Max Effort, Default Fallback) is the most intelligent AI model in 2026 with an AA Score of 57.6, according to the AA Intelligence Index by Artificial Analysis — a composite score updated every 6 hours. The language model (LLM) market in 2026 is dominated by a race between OpenAI, Anthropic, Google DeepMind, Meta AI and labs like DeepSeek, Alibaba (Qwen) and xAI (Grok). With over 875 models available via API, choosing the right model for each use case has become a complex decision involving quality (measured by benchmarks like AA Intelligence Index, MMLU and SWE-bench), token pricing, inference speed, context and multimodal capabilities.
Intelligence Index methodology — composite score composition. Sources: Artificial Analysis, LMArena, OpenRouter. Updated every 6h. CC BY 4.0.
AA Score — Artificial Analysis Intelligence Index↓
Intelligence Index (Artificial Analysis)↓
Pricing and Value↓
Context Window: Why It Matters↓
Open Source vs Proprietary↓
Frequently Asked Questions
What is the best AI model in 2026?↓
According to the AA Intelligence Index by Artificial Analysis, Claude Opus 5.5 (Adaptive Reasoning, Max Effort, Default Fallback) leads with a Score of 57.6. However, "best" depends on use case: for value, Gemma 4 E4B (Reasoning) offers excellent quality at low cost.
What is the AA Score?↓
The AA Score is the Intelligence Index by Artificial Analysis — a composite score (0-100) combining multiple reasoning, code, math and science benchmarks. It is automatically updated every 6 hours from the public Artificial Analysis API.
What is the cheapest AI model?↓
Among models with good quality (AA Score > 40), Claude Haiku 5.5 (Max, Default Fallback) is the most affordable at $0.10/1M input tokens.
Are benchmarks reliable?↓
Benchmarks are indicative, not definitive. The AA Intelligence Index is considered robust because it combines multiple standardized evaluations. Individual benchmarks (MMLU, GPQA) can suffer from contamination. We recommend testing on your specific use case.
ChatGPT vs Claude vs Gemini — which is better?↓
ChatGPT (OpenAI) is the most popular and versatile. Claude (Anthropic) excels at long reasoning and quality text. Gemini (Google) integrates well with the Google ecosystem. The ideal choice depends on your use case — use our comparator at /benchmark/comparar.