Models

Which AI model should you use?

186 models with live API prices, independent benchmarks and real usage. Filter by what you're building and see your monthly cost.

Prices tracked from the official pages of

  • OpenAI
  • Anthropic
  • Google
  • xAI
  • Mistral
  • DeepSeek
  • Perplexity
  • Groq

Compare AI models

Every model, ranked for your use case

Pick a use case, set your volume, and compare quality against what you'll actually pay.

What are you building?

Build software

Create media

Run in production

Voice & speech: Text-to-speech, realtime voice agents and speech-to-speech.

25 models · sorted by best fit

Share this view
xAI

Text to Speech

Balanced

Price

$15/ 1M chars

Standard

Voice & speech
Googlefree tier

Gemini 2.5 Flash Native Audio (Live API)

Balanced

Our Live API native audio models optimized for higher quality audio outputs with better pacing, voice naturalness, verbosity, and mood.

Input

$0.50

Cached input

—

Output

$2

per 1M tokens · Standard

Voice & speech
Googlefree tier

Gemini 3.8 Flash TTS

Balanced

Our 3.8 Flash Text-to-Speech audio model engineered for studio-grade voice fidelity, expressive acting, and long-form stability.

Input

$0.50

Cached input

—

Output

$9

per 1M tokens · Standard

Batch −50%Flex −50%Priority +80%
Voice & speech
Googlefree tierpreview

Gemini 2.5 Flash Preview TTS

Balanced

Our 2.5 Flash text-to-speech audio model optimized for price-performant, low-latency, controllable speech generation.

Input

$0.50

Cached input

—

Output

$10

per 1M tokens · Standard

Batch −50%
Voice & speech
OpenAI

tts-1

Balanced

Input

$15

Cached input

—

Output

—

per 1M tokens · Standard · Text

Voice & speech
Googlefree tierpreview

Gemini 3.1 Flash TTS Preview

Balanced

Our 3.1 Flash Text-to-Speech audio model optimized for price-performant, low-latency, controllable speech generation.

Input

$1

Cached input

—

Output

$20

per 1M tokens · Standard

Batch −50%
Voice & speech
Googlepreview

Gemini 2.5 Pro Preview TTS

Balanced

Our 2.5 Pro text-to-speech audio model optimized for powerful, low-latency speech generation for more natural outputs and easier to steer prompts.

Input

$1

Cached input

—

Output

$20

per 1M tokens · Standard

Batch −50%
Voice & speech
OpenAI

tts-1-hd

Balanced

Input

$30

Cached input

—

Output

—

per 1M tokens · Standard · Text

Voice & speech
OpenAI

gpt-realtime-2.1

Balanced

Input

$32

Cached input

$0.40

Output

$64

per 1M tokens · Standard · Audio

Voice & speech
OpenAI

gpt-realtime-2

Balanced

Input

$32

Cached input

$0.40

Output

$64

per 1M tokens · Standard · Audio

Voice & speech
OpenAI

gpt-realtime-1.5

Balanced

Input

$32

Cached input

$0.40

Output

$64

per 1M tokens · Standard · Audio

Voice & speech
OpenAI

gpt-realtime

Balanced

Input

$32

Cached input

$0.40

Output

$64

per 1M tokens · Standard · Audio

Voice & speech
OpenAI

gpt-audio-1.5

Balanced

Input

$32

Cached input

—

Output

$64

per 1M tokens · Standard · Audio

Voice & speech
OpenAI

gpt-audio

Balanced

Input

$32

Cached input

—

Output

$64

per 1M tokens · Standard · Audio

Voice & speech
Mistral

Voxtral TTS 3

Efficient

Price

$13/ 1M chars

Standard

Voice & speech
Mistral

Voxtral TTS

Efficient

Price

$16/ 1M chars

Standard

Voice & speech
Googlefree tier

Gemini 3.8 Flash-Lite TTS

Efficient

Our 3.8 Flash-Lite Text-to-Speech audio model optimized for high-throughput, low-latency, and cost-efficient conversational speech.

Input

$0.50

Cached input

—

Output

$6

per 1M tokens · Standard

Batch −50%Flex −50%Priority +80%
Voice & speech
OpenAI

gpt-4o-mini-tts

Efficient

Input

—

Cached input

—

Output

$12

per 1M tokens · Standard · Audio

Voice & speech
OpenAI

gpt-realtime-2.1-mini

Efficient

Input

$10

Cached input

$0.30

Output

$20

per 1M tokens · Standard · Audio

Voice & speech
OpenAI

gpt-realtime-mini

Efficient

Input

$10

Cached input

$0.30

Output

$20

per 1M tokens · Standard · Audio

Voice & speech

Not sure where to begin?

Answer 3 questions, get a model and a plan.

What you want to build, how much you code and your budget: you'll get the exact tools and the model to use.

Step 1 of 3

What are you building?

AI 101

New to AI? Six words that explain everything here

No jargon. Read this once and every page on AgentCost will make sense.

AI model

The “brain”. You send it text (or images, audio) and it sends back an answer. GPT, Claude, Gemini and DeepSeek are model families.

“Summarize this email” → a summary.

AI agent

A model plus tools and a loop: it plans, takes actions (edit files, browse, call APIs), checks results and keeps going until the task is done.

“Add a login page” → it writes code, runs it, fixes errors.

API

How your app talks to a model. You pay per use instead of a monthly seat, which is why your volume decides the bill.

Your support page sends each question to the model's API.

Tokens

The unit models are billed in — roughly ¾ of a word. Prices are “per 1M tokens”, with input (what you send) and output (what it writes) priced separately.

A 750-word article ≈ 1,000 tokens.

Context window

How much the model can read at once — its short-term memory. Bigger windows fit whole codebases or long documents in one request.

1M tokens ≈ several novels at once.

Benchmark

A standardized test that scores models on the same tasks, so you can compare quality fairly instead of trusting marketing.

Coding Index: how well a model solves real coding tasks.

Model or agent? The quick rule

You want an answer, text or a classification

→ A model

Call its API, or chat with it

You want a task done end-to-end

→ An agent

Use a ready agent or build one with a framework

You don't code and want a working app

→ An app-builder agent

Describe it, iterate by chatting

You want to cut cost

→ A smaller model

Use the “Best value” sort and test on real data

Usage trends

What developers are actually running

Real token volume across apps routed through OpenRouter (2026-10-29 – 2026-10-29). Not a survey — what production traffic is choosing right now.

Most-used models right now

Tokens processed in the latest ranking window.

  1. 1
    deepseek-v4.1-flashdeepseek
    58.3T
  2. 2
    glm-5.3-flashz-ai
    15.6T
  3. 3
    gpt-6-lunaopenai
    11.4T
  4. 4
    hy4-previewtencent
    9.2T
  5. 5
    claude-opus-5.5anthropic
    7.8T
  6. 6
    deepseek-v4-flashdeepseek
    6.8T
  7. 7
    step-5-previewstepfun
    5.9T
  8. 8
    glm-5.3z-ai
    4.9T
  9. 9
    minimax-m3.1-flashminimax
    4.1T
  10. 10
    nemotron-3-ultra-550b-a55bnvidia
    3.9T

Top AI apps by tokens this week

The tools developers build with every day.

  • H
    Hermes Agent14T tokens

    Hermes Agent is an open-source, self-improving AI agent by Nous Research that runs persistently with memory across sessions, and builds reusable skills from experience. It comes with 40+ built-in tools, including web search, browser automation, and vision, plus scheduled automations and subagents.

  • C
    Claude Code12.4T tokens

    Claude Code is Anthropic's agentic coding tool that reads your entire codebase, plans and executes changes across files, runs tests, and iterates on failures, all from natural language prompts.

  • P
    pi6.3T tokens

    There are many coding agents, but this one is yours.

  • C
    Codex5.6T tokens

    A coding agent that helps you build and ship with AI

  • K
    Kilo Code5T tokens

    Kilo Code is an open-source AI coding agent that works across VS Code, JetBrains, and CLI to help developers ship code faster with agentic workflows.

  • O
    OpenClaw3.2T tokens

    OpenClaw is an open-source AI agent that connects to your messaging apps and takes real actions on your behalf, from running commands and browsing the web to managing files and sending emails.

  • D
    DeepSeek Harness2.2T tokens
  • O
    omp2.1T tokens

Methodology

Sources, not opinions

Where do the prices come from?+

Straight from each provider's official pricing page, re-checked every few hours. Every model's “All prices” panel links the exact page it came from. If a page changes and our reader breaks, we keep showing the last verified prices and mark the provider below.

Where do the scores come from?+

We don't run benchmarks and never make up numbers. Intelligence, Coding and Agentic scores are Artificial Analysis's independent indexes; web and UI design scores are Design Arena's crowd-voted Elo ratings; “most used” ranks are real token volume from OpenRouter. Each is linked below.

How are “Our picks” chosen?+

From the models currently in view: the top performer has the highest score on the benchmark that fits your use case; best value has the most score per dollar among above-median models; the budget pick is the cheapest model that still scores above the median. Change filters or your workload and the picks update.

What do Frontier, Balanced and Efficient mean?+

A model's position in its own provider's lineup, read from the model family (Opus/Sonnet/Haiku, Pro/Flash/Lite, -mini…) and its price. It's a quick orientation aid — for actual quality, use the benchmark scores.

How is the monthly cost estimated?+

Your monthly input and output tokens × the model's per-million-token prices, with the cached share of input charged at the cache-read rate. It excludes tool fees, search grounding, images and taxes — always confirm on the provider's page before committing budget.

Is AgentCost affiliated with any AI provider?+

No. It's independent and free to use. No provider pays for placement or rankings.

Benchmark & usage sources

  • Artificial Analysis

    Independent Intelligence, Coding and Agentic indexes

  • Design Arena

    Crowd-voted Elo for websites, UI components and other design tasks

  • OpenRouter

    Real token usage across apps routed through OpenRouter

Pricing data status

Community

Built for builders, shaped by them

Every testimonial is verified by hand before it's published.

Reach the developer

Built by one developer. Every message gets read.

Spotted a wrong price, want a provider added, or have an idea that would make this more useful for your team? Partnerships and collaborations are welcome too.

Connect on LinkedIn