PROJECT 03 / BROWSER TOOL
Visionovisor
Explore AI models, compare API costs, understand model capacity, and count document tokens in your browser.
Meet the models
Explore text, coding, small language, embedding and reranking models. Each profile shows its developer, purpose and availability.
No matching models. Try another name or model type.
USD API estimates; these are separate from ChatGPT, Claude and other subscription fees. Tool charges, images, audio, taxes, caching discounts and batch discounts are excluded.
Pricing rules and calculation
Input tokens ÷ 1,000,000 × input rate + billed output tokens ÷ 1,000,000 × output rate. Multiply the unrounded per-request cost by the number of requests. Rates may depend on the complete input length. Instructions are a capacity reservation; include their billed tokens in the input figure yourself when estimating a complete request.
Capacity is based on tokens and provider limits. Parameter counts do not determine document capacity. English word/page estimates are adjustable below. Reasoning can reduce the visible answer.
Count a document privately
TXT, DOCX or PDF with extractable text · maximum 10 MiB
| Model | Text tokens | Usable input | Occupied | Request budget |
|---|
Counts cover extracted text only. They exclude native PDF/image processing, conversation history and hidden system/tool overhead. Partial extraction cannot assess the full document.
Supported tokenizers and document limitations
Other models remain available for costs and information. A tokenizer is enabled only after its exact model version and local implementation have been verified. No character division or approximate word/token ratio is substituted. Files remain in memory and are cleared on reset or when you leave this page. No API keys, document uploads, analytics or document-content logging are used.
Scanned pages require OCR, which is outside this version. PDF pages are reported from the file; DOCX pagination and word-based pages are estimates. Complex layout, inaccessible PDF fonts and omitted embedded objects may affect extraction. Text extraction is not visual verification of every page.
Capacity and English-text assumptions
Usable input reserves instructions and the expected answer within shared context. Separate input/output limits are respected without subtracting output twice. These English estimates are not reliable for Arabic, Kurdish, code, tables or images. Answer capacity is capped by the output limit and remaining context; it is never the entire advertised context window.