Vibedia.
What the models cost, and which one to use.
Your vendors. Tap one to follow it across the whole site.
Following some vendors.Everything below is filtered to them. Tap again to unfollow, or clear it in the header.
Start here. Pick the job, see which models can do it and what a run costs.
Build a websiteGPT-5 Nano$0.1860Gemini 2.5 Flash Lite$0.2760GPT-4.1 nano$0.2760from $0.1860 a runChange code that already existsGPT-5 Nano$0.1300Gemini 2.5 Flash Lite$0.2300Claude Haiku 5.5$0.2375from $0.1300 a runWrite and reply to emailGPT-5 Nano$0.00022Gemini 2.5 Flash Lite$0.00028GPT-4.1 nano$0.00028from $0.00022 a runMake a presentationGPT-5 Nano$0.00600Gemini 2.5 Flash Lite$0.00780GPT-4.1 nano$0.00780from $0.00600 a runWork on a spreadsheetGPT-5 Nano$0.00352Gemini 2.5 Flash Lite$0.00512Claude Haiku 5.5$0.00560from $0.00352 a runSummarise long documentsGPT-5 Nano$0.00340Gemini 2.5 Flash Lite$0.00640GPT-4.1 nano$0.00640from $0.00340 a runResearch a topicGPT-5 Nano$0.0180Gemini 2.5 Flash Lite$0.0280Claude Haiku 5.5$0.0300from $0.0180 a runReview codeGPT-5 Nano$0.00160Gemini 2.5 Flash Lite$0.00260Claude Haiku 5.5$0.00275from $0.00160 a run
The latest. Collected every three hours, with where each item came from.
OpenAIAdded to the catalogue: gpt-rosalind-research at 5.0 in and 25.0 out per million tokens2026-10-10 · found by the price refreshOpenAIAdded to the catalogue: gpt-rosalind-discovery at 5.0 in and 25.0 out per million tokens2026-10-10 · found by the price refreshOpenAIAdded to the catalogue: gpt-5.6-cyber at 12.5 in and 75.0 out per million tokens2026-10-10 · found by the price refreshAI21Added to the catalogue: Jamba Mini at 0.2 in and 0.4 out per million tokens2026-10-10 · found by the price refreshAI21Added to the catalogue: Jamba Large at 2.0 in and 8.0 out per million tokens2026-10-10 · found by the price refreshxAIAdded to the catalogue: Grok 4.20 (Non-Reasoning) at 1.25 in and 2.5 out per million tokens2026-10-10 · found by the price refreshGoogleAdded to the catalogue: Gemini Omni Flash Preview at 1.5 in and 17.5 out per million tokens2026-10-10 · found by the price refreshGoogleAdded to the catalogue: Gemini 3.1 Flash Live Preview at 0.75 in and 4.5 out per million tokens2026-10-10 · found by the price refreshGoogleAdded to the catalogue: Gemini 3 Flash Preview at 0.5 in and 3.0 out per million tokens2026-10-10 · found by the price refreshOpenAIAdded to the catalogue: GPT-5.6 at 4.0 in and 20.0 out per million tokens2026-10-10 · found by the price refresh
Prompts. 19 that say which line is doing the work.
codeWork out why it is brokenclaude-code · cursor · cli · chatcodeUnderstand code before changing itclaude-code · cursor · idecodeReview my change before I ship itclaude-code · cursor · ide · clicodeWrite tests for code that already worksclaude-code · cursor · ideemailWrite the email I am avoidingchatemailReply to an email in my voicechat · copilotlinkedinWrite a LinkedIn post in my actual voicechatlinkedinCut a post down without losing the argumentchat
The words. Defined where they appear, not in a separate dictionary.
Worth knowing
Cache write
The charge for putting a prefix into the prompt cache. Usually more than the input rate, not less.
An agent re-sends its context each turn, so theprompt cache (paying a reduced rate for a prefix the vendor has already processed) is what keeps the bill down — and the write is what puts it back up.
That is how every term on this site reads with Explain jargonswitched on in the header. Off, it is a quiet dotted underline.
Checked, not collected. Every figure names the page it was read from.
EvidenceVendor pageCorroboratedOne sourceSources disagree50 confirmed on a vendor pageNever zeronot publishedWhat a cell says when a vendor publishes no figure. Zero would read as free.50 publish no cache-write priceSources26 of 27Answered when we last looked. The ones that did not are named.A silent gap is a lieHeld back219Listed by one database and not confirmed, so they are published separately rather than mixed into the table.Not prices we will stand behindThe spread600×Between GPT-5 Nano at $0.05 and GPT-5.4 Pro at $30, per million input tokens.64 models · 7 vendorsTake itmodels.jsonThe whole catalogue, every figure with the URL it came from. Free to reuse.Updated 2026-10-10
Understand it. 7 lessons, 5 craft guides, about an hour.
Lesson 1What a language model actually isA next-token predictor, and why that description is both accurate and far too small.Read in orderLesson 2Tokens, and why everything is counted in themThe unit everything is counted and charged in, and why it is not words.Read in orderLesson 3Generative, agentic, and the gap between themThe difference between a model that writes and a model that acts, and why it matters for your bill.Read in orderLesson 4What it costs, and why the surpriseWhere the money actually goes, which is rarely where people expect.Read in orderPlaybookRunning an agent without a surprise billThe context grows on every pass, so the last turn costs several times the first.Comes with a widgetPlaybookChoosing a modelThere is no best model; there is a best model for a task at a price.Comes with a widgetPlaybookWorking inside a context windowA context window is a budget spent on every call, not a memory that accumulates.Comes with a widgetPlaybookPrompt caching, and when it costs you moneyReading from the cache is cheap; writing to it costs more than sending normally.Comes with a widget
Prices checked 2026-10-10. Newsroom checked 2026-10-10.How old everything is.