Measure the workload
the API will actually see.
Original, source-linked explanations of privacy-first multimodal token accounting: confidential prompts, extracted document text, source files, image geometry, context fit, and the boundary between a browser-local count and a provider invoice.
OpenAI Dots: what an always-on agent changes
Dots are GPT-6 Astra agents that keep working after you close the tab. What they can do, and why scheduled work changes the token footprint.
Grok Bot: one computer, many screens
xAI gives every Bot a screen on one shared cloud computer. How that architecture works, and why the 200K pricing cliff hits agents and never hits chats.
Meta Muse: the agent that meters itself in tokens
500 million Muse tokens a week for $20. What that number means, what a Muse token is not, and the contributor tier that trades your prompts for a 21x discount.
Every scheduled 2027 API price increase, in one table
Three Gemini Flash models and both Gemini TTS models double on 1 January 2027, while Anthropic cancelled a planned rise and OpenAI halved two tiers.
Claude Sonnet 4.5 retires on 30 November: what the migration really costs
The replacement costs 33 percent less per token but counts about 30 percent more tokens, so the real saving is about 13 percent. With a migration checklist.
Jev and the cost of a decision: a decision model against reasoning LLMs
TypeSafe's Jev bills $0.042 per million input tokens and nothing for output. What that does to the cost of routing and classification against Claude and OpenAI models.
DeepSeek V4.1 Flash: the price, the peak hours, and what changed
Off-peak and peak prices for V4.1 Flash and V4 Pro, when peak hours fall in China, Europe and the US, and what a mixed traffic pattern really pays.
Muse Spark 1.3: what it costs, and what the Contributor tier costs you
Standard and Contributor pricing for Meta's Muse Spark 1.3, a worked monthly example, and independent scores against Claude Opus 5.
Private LLM token counting for enterprise prompts, files, and code
A verifiable browser-local architecture for preflighting confidential AI workloads without uploading their contents to a counting service.
How to calculate tokens in PDFs, DOCX, code, images, and media
A local, multimodal workflow for measuring everything that enters an LLM request: extracted text, page images, vision tokens, audio seconds, and video frames.
Image token calculation across OpenAI, Claude, and Gemini
The same 1920x1080 screenshot costed on five models: OpenAI 32px patches, Claude 28px patches in two tiers, and Gemini 3 fixed budgets.
Why a token count is not the same as an API bill
A technical boundary map for message framing, tools, caching, reasoning, output, and long-context pricing.
Every guide identifies calculation boundaries, links to primary provider documentation, carries a visible review date, and is updated only when the underlying method or source changes.