OpenAI
gpt-5-search-api
Input
$1.25
Cached input
$0.125
Output
$10
per 1M tokens · Standard
Est. monthly cost$291.25
Cheapest in view
Backend developmentFrontend developmentAgents & automationChatbots & supportResearch & web search
Live prices · independent benchmarks · real usage
Build anything with AI.Know what to use.
Whether you've never written code or ship to production daily: AgentCost tells you which AI model or agent fits your idea, what it really costs, and what other builders are using right now.
- AI models
- 180
- Agents & tools
- 33
- With benchmarks
- 75
- Prices checked
- 6h ago
Live data
Top models by Coding Index
Scores: Artificial Analysis · Prices per 1M tokens (in / out)
Prices tracked from the official pages of
- OpenAI
- Anthropic
- xAI
- Mistral
- Perplexity
- Groq
Start here
Not sure where to begin? Answer 3 questions.
Tell us what you want to build, how much you code and your budget. You'll get a step-by-step plan with 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
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
Research & web search: Answers grounded in live web results, with citations.
Estimate your monthly bill
Every token-priced card shows its cost for this workload.
4 models · sorted by best fit
Share this view
Balanced
OpenAI
gpt-rosalind-research
Input
$5
Cached input
$0.50
Output
$25
per 1M tokens · Standard
Est. monthly cost$865
3.0× the cheapest in view
Backend developmentFrontend developmentAgents & automationChatbots & supportResearch & web search
Perplexity
Search API with Fast Search
Lower-latency web search results. Set search_type: "fast"
Price
$0.001/ request
Standard
Research & web search
Perplexity
Search API
Raw web search results with advanced filtering
Price
$0.005/ request
Standard
Research & web search
AI agents & tools · live prices & stats
Agents that build, code and automate for you
Models answer; agents get things done. Plans and prices are read live from each tool's pricing page, popularity from GitHub and real usage — click any tool for the full breakdown.
AI app builders
No coding neededDescribe it, get a working app
You type what you want (“a booking site for my salon”) and the agent writes the code, shows a live preview and can publish it. You refine it by chatting.
Best for: Non-coders, founders validating an idea, quick MVPs and landing pages.
- LLovableChat-to-app builder for full web apps with hosting.Pricing on site
- BBoltPrompt, run, edit and deploy full-stack apps in the browser.Free plan · from $18/mo
- vv0Vercel's AI builder for web apps and UI, deployable to Vercel.Free plan · from $30/user/mo
- RReplit AgentBuilds and hosts apps inside Replit's online workspace.from $20/mo
Automation platforms
No coding neededConnect your apps, let AI do the busywork
Visual workflows (“when an email arrives, summarize it with AI and post to Slack”). You drag blocks instead of writing code, and plug an AI model into any step.
Best for: Business owners and teams automating repetitive tasks across tools they already use.
AI code editors
Some coding helpsWrite code with an AI pair-programmer
A code editor with AI built in: it completes your code, explains errors and can make multi-file changes when you ask. You stay in control of every change.
Best for: People learning to code and developers who want speed without giving up control.
- CCursorAI-first code editor built on VS Code, with an agent mode.Free plan · from $8/mo
- GGitHub CopilotAI assistant inside VS Code, JetBrains and GitHub.Free plan · from $10/mo
- CCline#7 trendingOpen sourceOpen-source coding agent that runs inside your IDE.Free · Enterprise custom★ 70.9k
- KKilo Code#3 trendingOpen sourceOpen-source coding agent for VS Code, JetBrains and the CLI.Free plan · from $15/user/mo★ 27.8k
Autonomous coding agents
For developersHand off whole tasks, review the result
Agents that read your whole project, plan, edit many files, run tests and fix failures on their own — usually from the terminal. You review what they did, like a teammate’s pull request.
Best for: Developers shipping features, refactors and bug fixes faster.
- CClaude Code#1 trendingAnthropic’s agentic coding tool for the terminal and IDE (included in Claude plans).Free plan · from $17/mo★ 153k
- CCodex#6 trendingOpen sourceOpenAI’s coding agent for the CLI, IDE and cloud (included in ChatGPT plans).Pricing on site★ 131k
- GGemini CLIOpen sourceGoogle’s open-source coding agent for the terminal.★ 108k
- OOpenHands#13 trendingOpen sourceOpen-source agent that runs commands, browses and edits code.★ 91.9k
Agent frameworks
For developersBuild your own AI agent into your product
Code libraries for building agents yourself: you pick a model, give it tools (search, your database, your APIs) and the framework runs the reasoning loop.
Best for: Developers adding AI assistants, copilots or automated workflows to their own app.
- OOpenAI Agents SDKOpen sourceLightweight Python framework for multi-agent workflows.★ 30.3k
- CClaude Agent SDKOpen sourceThe agent loop behind Claude Code, as a library.★ 8.4k
- LLangGraphOpen sourceGraph-based framework for stateful, controllable agents.★ 43.9k
- CCrewAIOpen sourceFramework for teams of role-based agents working together.Free · Enterprise custom★ 60.1k
- VVercel AI SDKOpen sourceTypeScript toolkit for AI features and agents in web apps.★ 27.5k
- MMastraOpen sourceTypeScript agent framework with workflows, memory and evals.★ 29.1k
Prices are read live from each tool's pricing page, stars and releases from GitHub, usage from OpenRouter. Click any tool for plans and details.
Usage trends
What developers are actually running
Real token volume across apps routed through OpenRouter (2026-10-22 – 2026-10-25). Not a survey — what production traffic is choosing right now.
Most-used models right now
Tokens processed in the latest ranking window.
- 151.6Tdeepseek-v4.1-flashdeepseek
- 213.7Tminimax-m3.1-flashminimax
- 313.2Tglm-5.3-flashz-ai
- 49Thy4-previewtencent
- 58.4Tgpt-6-lunaopenai
- 68.3Tclaude-opus-5.5anthropic
- 76.8Tdeepseek-v4-flashdeepseek
- 86.1Tmimo-v2.6-flashxiaomi
- 95.9Tnemotron-3-ultra-550b-a55bnvidia
- 105Tglm-5.3z-ai
Top AI apps by tokens this week
The tools developers build with every day.
- CClaude Code13T 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.
- HHermes Agent12.9T 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.
- KKilo Code6.5T 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.
- Ppi5.9T tokens
There are many coding agents, but this one is yours.
- DDeepSeek Harness2.9T tokens
- Oomp2.6T tokens
- CCodex2.1T tokens
A coding agent that helps you build and ship with AI
- FFreebuff2T tokens
Powerful coding models, funded by ads
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
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.
Stop overpaying for AI.
The model that's good enough for your use case can be 10× cheaper than the one you default to. Find yours in a minute.