Context · Strategic Landscape

State of the AI Market

678 models · 68 providers

The live index spans 678 models from 68 providers across frontier, reasoning, efficient, multimodal, and embedding categories. Across the analytical sample of 668 records with published input pricing, the average blended price is $3.385 per million tokens (median $0.800), and the largest context window reaches 2M tokens. How this is calculated →

The Landscape

Explore the AI landscape

API ProvidersEvery major LLM provider, with live pricing and model rosters.68 providers Company MapChipmakers, labs, clouds, and datacenters across the AI economy.47 companies Who's Who in AIThe founders, CEOs, and researchers shaping the field.34 profiles AI GlossaryPlain-language definitions of the terms behind AI pricing.63 terms

Market Snapshot

The market at a glance

Models Tracked

678

Canonical index

Providers Tracked

Canonical index

Avg Blended Cost

$3.385/M

668 priced records

Median Blended

$0.800/M

Analytical-sample midpoint

Max Context

2M tokens

In analytical sample

Free-tier Models

In analytical sample

Average cost by model category

Analytical sample · 5 categories

Frontier45 sample models

avg $2.265/Mmedian $1.520/M

Reasoning29 sample models

avg $1.137/Mmedian $1.042/M

Efficient325 sample models

avg $1.182/Mmedian $0.400/M

Multimodal270 sample models

avg $6.600/Mmedian $1.625/M

Embedding9 sample models

avg $0.508/Mmedian $0.100/M

Provider pricing comparison

Analytical sample · Top 10 by avg blended cost

OpenAI100 sample models

avg $12.574/M blended

Anthropic33 sample models

avg $11.157/M blended

sakana2 sample models

avg $7.183/M blended

Inflection AI3 sample models

avg $4.750/M blended

Perplexity7 sample models

avg $4.143/M blended

Zhipu AI4 sample models

avg $3.710/M blended

Reka AI3 sample models

avg $3.080/M blended

xAI22 sample models

avg $2.955/M blended

AI21 Labs4 sample models

avg $2.105/M blended

Replicate5 sample models

avg $2.033/M blended

Model distribution by category

Analytical sample

Efficient

325

Multimodal

270

Frontier

Reasoning

Embedding

AI Provider Intelligence

Provider intelligence

Analytical sample · 68 providers represented

efficientfrontiermultimodal

efficientfrontiermultimodal

Llama 3.1 70B (Anyscale)

ERNIE 4.5 VL 424B A47B

efficientmultimodalreasoning

frontierefficientmultimodal

Llama-3.3-70B (Cerebras)

efficientreasoning

View all models

cognitivecomputations

reasoningefficientmultimodalfrontier

Dots3-Note Preview (free)

multimodalfrontierefficientembed

efficientmultimodalreasoning

Llama 3.1 70B (Lepton)

LiquidAI: LFM2.5-2.6B (free)

multimodalfrontierefficientembed

efficientreasoningmultimodal

Llama-3.1-Nemotron-Ultra-253B

efficientfrontiermultimodal

multimodalreasoningembedefficientfrontier

multimodalefficientreasoning

efficientfrontiermultimodalreasoning

Llama 3.3 Euryale 70B

Thinking Machines: Inkling (batch)

Qwen2.5-72B (Together)

frontierefficientembed

Companies shaping AI

Full company map →

AI21 LabsIsraeli lab building Jamba hybrid SSM-Transformer models and reliability-focused enterprise AI systems.AlibabaChinese tech giant behind Qwen, the world's most widely adopted open-weight model family.AmazonAWS runs the largest cloud, backs Anthropic, and builds Trainium chips to challenge NVIDIA.AMDNVIDIA's primary GPU challenger with the Instinct MI line, plus the EPYC CPUs powering AI datacenters.Andreessen HorowitzThe venture firm (a16z) backing a large share of the most prominent AI model and application startups.AnthropicAI safety-focused lab behind the Claude model family, backed by Amazon, Google, and enterprise adoption.AnysphereMaker of Cursor, the AI code editor that became one of the fastest-growing developer tools ever.ArmThe chip-architecture licensor whose energy-efficient designs are spreading into AI data centers.BroadcomThe quiet giant co-designing custom AI accelerators for hyperscalers and supplying AI networking silicon.CerebrasBuilder of the wafer-scale engine — the largest chip ever made — for fast AI training and inference.Character.AIConsumer platform for conversational AI characters, whose founders returned to Google in 2024.CohereEnterprise-focused model builder specializing in secure, privately deployable AI for business and government.

Who's Who in AI

All profiles →

Aidan GomezCEO, Cohere Alexandr WangChief AI Officer, Meta Andrej KarpathyFounder, Eureka Labs Andrew FeldmanCEO, Cerebras Andy JassyCEO, Amazon Aravind SrinivasCEO, Perplexity Arthur MenschCEO, Mistral AI Brett AdcockCEO, Figure

FAQ

Frequently asked questions

What is the average AI model price?

Across the current analytical sample of 668 model records with published input pricing, the average blended price is $3.385 per million tokens, with a median of $0.800 per million. When an output rate is unavailable, the blended calculation uses the input rate for both sides.

Which model category is most expensive?

Within the analytical pricing sample, Multimodal models are the most expensive category, averaging $6.600 per million tokens blended. Efficient and embedding models sit at the low end, often a fraction of a cent per million.

How many AI providers are there?

We currently track 68 providers, including frontier labs and inference platforms, spanning 678 distinct models across frontier, reasoning, efficient, multimodal, and embedding categories.

What is the largest context window available?

The largest context window in the current analytical sample is 2M tokens. Context windows determine how much text a model can read and reason over in a single request, and they vary widely between providers.

Quarterly Report

Q2 2026 · State of AI Infrastructure

The definitive benchmark for enterprise AI spending: pricing trends, provider analysis, and market forecasts for AI infrastructure decision-makers. View the live pricing index →

Free access during Phase 1. No credit card required.