PriceIndex

Labs & inference hosts

AI Providers

Compare 110 AI providers side by side — who builds each model, what their APIs cost per million tokens, how large their context windows are and how they score on benchmarks. Use it to pick the right AI vendor before you commit.

110

Providers tracked

20 labs · 90 hosts

317

Models priced

Input, output & cached rates

$0.02

Cheapest model

Mistral Nemo · Mistral AI

90

Top quality score

GPT-5.2 Pro · OpenAI

At a glance

AI provider comparison table

Every provider's model count, cheapest blended price (75% input / 25% output per 1M tokens), largest context window, best benchmark score and 6-month price trend. Click any column to sort.

Google logoGooglelab
Models
27
Cheapest /1M
$0.06
Avg blended
$1.66
Max context
1M
Top Arena Elo
82
6-mo trend
-2%
Anthropic logoAnthropiclab
Models
17
Cheapest /1M
$0.20
Avg blended
$9.22
Max context
1M
Top Arena Elo
57.6
6-mo trend
0%
OpenAI logoOpenAIlab
Models
54
Cheapest /1M
$0.04
Avg blended
$8.22
Max context
1.1M
Top Arena Elo
90
6-mo trend
1%
XAxAIlab
Models
7
Cheapest /1M
$1.25
Avg blended
$2.13
Max context
2M
Top Arena Elo
46.4
6-mo trend
23%
Meta logoMetalab
Models
14
Cheapest /1M
$0.06
Avg blended
$0.59
Max context
1.3M
Top Arena Elo
48.1
6-mo trend
9%
DeepSeek logoDeepSeeklab
Models
15
Cheapest /1M
$0.04
Avg blended
$0.50
Max context
1M
Top Arena Elo
39.5
6-mo trend
32%

6 of 110

Visualised

AI provider price comparison charts

How average and entry-level prices differ between providers, and which vendors give the most quality per dollar.

Average vs cheapest price by provider

Blended USD per 1M tokens, excluding embedding models.

Price vs benchmark, every model

Up and to the left means better value. Price axis is logarithmic.

#1OpenAI logo

OpenAI

Best: GPT-5.2 Pro · score 90

#2Google logo

Google

Best: Nano Banana Pro (Gemini 3 Pro Image) · score 82

#3Anthropic logo

Anthropic

Best: Claude Opus 5.5 · score 57.6

Directory

Every AI provider and its models

Logos, positioning, key numbers and the full model lineup for each provider — with prices linked to detailed model pages.

Google logo

Google

lab

Gemini multimodal models via AI Studio and Vertex.

Models
27
From /1M
$0.05
Max context
1M
Google pricing & models
Anthropic logo

Anthropic

lab

Safety-focused lab behind the Claude family.

Models
17
From /1M
$0.10
Max context
1M
Anthropic pricing & models
OpenAI logo

OpenAI

lab

Creator of GPT and o-series reasoning models.

Models
54
From /1M
$0.02
Max context
1.1M
OpenAI pricing & models
Meta logo

Meta

lab

Open-weight Llama models, hosted by many providers.

Models
14
From /1M
$0.03
Max context
1.3M
Meta pricing & models
DeepSeek logo

DeepSeek

lab

Aggressively priced open reasoning models.

Models
15
From /1M
$0.02
Max context
1M
DeepSeek pricing & models
Mistral AI logo

Mistral AI

lab

European lab with open and commercial models.

Models
20
From /1M
$0.02
Max context
1M
Mistral AI pricing & models
QW

Qwen

lab

Models
53
From /1M
$0.03
Max context
1M
Qwen pricing & models
Z.

Z.ai

lab

Models
16
From /1M
$0.04
Max context
1M
Z.ai pricing & models
AI

AI21

host

AI21 Labs is an Israeli company specializing in natural language processing (NLP), developing advanced AI systems and foundation models designed for enterprise applications. Their mission is to empower businesses with state-of-the-art Large Language Models (LLMs) and AI applications to redefine how humans interact with text.

Models
0
From /1M
—
Max context
—

Hosts open-weight models such as Llama 3.3 70B. Dedicated pricing coming soon.

AI21 pricing & models

AI labs: who builds the models

Labs train foundation models and sell first-party API access. Pricing is set by the lab, and new models and price cuts usually appear here first.

GoogleAnthropicOpenAIxAIMetaDeepSeekMistral AIQwenZ.aiMiniMaxMoonshotAITencentAionLabsByteDance SeedCohereNVIDIAPerplexityXiaomiMicrosoftVoyage AI

Inference hosts: who serves them

Hosts run open-weight models such as Llama or DeepSeek on their own hardware. The same model can cost very different amounts, and run at different speeds, depending on the host.

AlibabaSakana AIStepFunUpstageInceptionInferenceNetMorphNex AGIPerceptronPoolsideRelaceThinking MachinesUnbiasedArcee AIBaiduFireworksMancer 2VeniceAI21AkashMLAmazon BedrockAmazon NovaAmbientAssemblyAIAtlasCloudAvianAzureBaseTenBlack Forest LabsCerebrasChutesCirrascaleClarifaiClaude Platform on AWSCosineCrucibleCrusoeDarkbloomDatabricksDecartDeepgramDeepInfraDekaLLMDigitalOceanFakeProviderFeatherlessFish AudioFriendliGMICloudGoogle AI StudioGroqHeyGenInceptronInferact vLLMInfermaticInflectionIo NetIonstreamKreaLiquidMakoraMaraModelRunModularNear AINebiusNextBitNovitaOllamaOpenInferenceParasailPhalaPrimeIntellectQuiverRecraftRekaRunwaySail ResearchSambaNovaSeedSiliconFlowSourcefulStealthStreamLakeSwitchpointTenstorrentTogether AITypeSafeVoyageAI by MongoDBWafer

Head to head

Popular cross-provider comparisons

Flagship models from rival providers, compared on price, context and benchmark.

Nano Banana Pro (Gemini 3 Pro Image) vs Claude Opus 5.5Nano Banana Pro (Gemini 3 Pro Image) vs GPT-5.2 ProNano Banana Pro (Gemini 3 Pro Image) vs Grok 4.7Nano Banana Pro (Gemini 3 Pro Image) vs Muse Spark 1.3Nano Banana Pro (Gemini 3 Pro Image) vs DeepSeek V4.1 FlashNano Banana Pro (Gemini 3 Pro Image) vs Mistral Medium 3.5Nano Banana Pro (Gemini 3 Pro Image) vs Qwen3.8 2.4T A95BNano Banana Pro (Gemini 3 Pro Image) vs GLM 5.3Nano Banana Pro (Gemini 3 Pro Image) vs MiniMax M3

FAQ

AI provider questions, answered

Which AI provider is the cheapest?

Right now the lowest blended price we track is Mistral Nemo from Mistral AI at $0.02 per 1M tokens. Open-weight models served by hosts are often the cheapest route for high-volume workloads.

Which AI provider has the best models?

GPT-5.2 Pro from OpenAI has the highest measured quality score in our dataset (quality score 90 · Arena Elo 1255 · MMLU-Pro 82.8%). The best choice still depends on your task, latency and budget.

Which provider offers the largest context window?

Grok 4.20 from xAI supports up to 2M tokens.

What is the difference between an AI lab and an inference host?

A lab trains and publishes its own models; an inference host runs models (usually open-weight ones) on its infrastructure and sells API access to them.

How is the blended price calculated?

Blended price assumes a typical 3:1 input-to-output token ratio: 75% of the input price plus 25% of the output price, per 1M tokens.

How often are provider prices updated?

Prices are checked against official pricing pages and every change is stored with its date, so the history charts show exactly when a provider changed its rates. Always confirm with the vendor before committing.