Home / Guides
Guides: what AI costs in practice
Explainers built on our daily price data: tokens, reasoning, caching, plans vs API, and the real cost of common jobs.
ChatGPT Plus vs Business: which plan is worth it?Plus costs far less per seat than Business, but the gap is about more than price. Limits and team features decide which one fits your work.Gemini vs DeepSeek API: which is cheaper per job?Google and DeepSeek both push low prices for high-volume work. Their cheapest models behave differently on writing and data tasks.DeepSeek vs Claude API: cost per task comparedDeepSeek is often the cheapest name on a pricing page, but Claude models are chosen for different reasons. The real question is what each onOpenAI vs Anthropic API: which costs less per task?Both companies sell several model tiers, and the cheapest per token is not always the cheapest per finished job. This piece compares matchinWhat does an AI code review cost per pull request?Reviewing 497 lines of real Python costs from a fraction of a cent to a few cents. Measured in 12 models.How much does an AI coding agent cost per task?I modeled a 50-step agent run in every model. The spread is huge.How much does an AI customer support bot cost?Price a support bot by the ticket: triage, answer, reply email. Real per-task costs.AI price history: how LLM prices changed from 2024 to 2026I rebuilt two and a half years of list prices from archived copies of the OpenRouter model list.AI prices are not only falling: the 2026 price increasesSeveral models got more expensive after launch. Here is the list from my price log.AI pricing glossary: 25 terms on your AI bill, explainedCached input, reasoning effort, context window, batch API and more, in plain words with real numbers.The reasoning trap: models that burned 16,000 tokens on a spreadsheetI asked 12 models to analyze a 200-row CSV. Several thought until they hit my 16,000-token cap.Every AI subscription compared: ChatGPT, Claude, Gemini, Grok and more49 paid AI plans in one place, with what each costs per month and what it includes.AI video generation cost per minute: Veo, Kling, Seedance, Wan and moreOne minute of 1080p AI video costs from about five dollars to about seventy, depending on the model and the provider.How much does it cost to write a blog post with AI?A 1,500-word article costs a few cents at most. The surprise is how much the models disagree on length.How much does the ChatGPT API cost per message?One chatbot answer on the OpenAI API costs a fraction of a cent, or a few cents with the top model. Measured numbers inside.ChatGPT Plus vs the API: which is cheaper for you?Twenty dollars a month buys a lot of API calls. Here is the break-even in real chat answers.Cheapest AI for translation: cost per 1,000 wordsTranslating 1,000 words with an LLM costs well under a cent in most models.Cheapest AI image generator API: cost per 1K and 4K imageA 1K image costs from under a cent; a 4K image from about ten cents. Every official price, compared.The cheapest LLM APIs in October 2026, by real task costCheapest per token is not cheapest per task. My ranking by what jobs actually cost.Claude API cost per request: Haiku, Sonnet, Opus and Fable comparedWhat one request to Claude actually costs, from Haiku 5.5 to Fable 5.1, on real tasks.How much does Claude Code cost per month? Plans vs APIClaude Code runs on a Claude plan or on API credit. The modeled cost of an agent run tells you which is cheaper for you.Claude Pro vs Max vs the API: what you get for the moneyPro, Max 5x, Max 20x or pay per token. A comparison with real per-request costs.Claude counts about 39% more tokens for the same text. Here is what it costsI sent the same 10,000-word document to every lab. The Anthropic tokenizer counted the most tokens.What a RAG query costs: retrieval-augmented answers pricedFive retrieved documents plus a question: what one grounded answer costs.How much does it cost to summarize documents with AI?A 20-page document summary costs from a fraction of a cent to over 20 cents. Here is why.DeepSeek pricing explained: cheap per token, not always per taskDeepSeek V4.1 Flash has low list prices, but it thinks a lot. What that means for your bill.How to estimate your monthly AI API billCount tasks, not tokens. A worked example and a calculator that does the math for every model.Gemini API pricing explained per taskGemini 3.8 Flash, 3.5 Flash-Lite and 3.1 Pro, priced by the job instead of by the token.GitHub Copilot vs Cursor: what AI coding plans really costPlan prices, included credits and what the same money buys on the API.GPT-6.1 Sol vs Claude Sonnet 5.5: which costs less per task?Same list price per token, different bills. Task by task.How to reduce LLM API costs: 10 techniques with real savingsModel choice, caching, batch, reasoning effort and shorter outputs. What each one saves on real tasks.Input vs output tokens: why output costs moreOutput tokens cost several times more than input tokens. Which tasks that hurts, and which it does not.Open-weight vs closed AI models: which is cheaper per task?Open models are cheap per token, but not always per task. A measured comparison.Prompt caching: how much it saves on AI agentsA 50-step agent re-reads its whole context at every step. Caching is the difference between cents and dollars.Why the same AI model costs different amounts at different providersOpen models are served by many hosts at different prices. Closed models usually cost the same everywhere.Text-to-speech cost per minute: ElevenLabs vs OpenAI vs Google and moreOne minute of AI voice is about 900 characters. Here is what that costs in every TTS API.Tokens to words: how many tokens is 1,000 words?Rules of thumb are fine until the bill arrives. I counted tokens for the same texts in every major model family.Transcription cost per hour: Whisper, Deepgram, AssemblyAI, Google and moreTranscribing an hour of audio costs from a few cents to over a dollar. Official prices, compared.What is a token? How AI pricing really worksA token is the unit AI APIs bill by. Here is how many tokens real tasks use and why the same text costs more in some models.Why reasoning models cost more: the hidden thinking tokensYou pay for tokens you never see. On some tasks the thinking is ten times longer than the answer.