Guides
Practical, technical guides for everything tokens. Each guide answers a specific question with worked examples and real numbers from the same pricing data that powers the live counter. Nothing here is padded: if a question has a two-sentence answer, the guide is two sentences plus the math that backs it up.
How many tokens are in...
The estimation guides. Useful when you need a budget number before you have the actual text, or when the input is not text at all.
- How many tokens are in a PDF?, page-level estimates for text PDFs, image PDFs, and the OCR catch
- How many tokens are in a word?, the rough heuristic and where it breaks
- How many tokens are in an image?, multimodal token math for GPT-4o, Claude, and Gemini vision inputs
- How many tokens are in a book?, worked estimates from short novels to multi-volume works
- How many tokens per page?, varies by font size and content density; here are the numbers
Cheapest model for your workload
Different workload shapes have different cost winners. A model that wins on short chat turns can lose badly on input-heavy RAG, because input and output prices differ by 3-5x on most models. Each ranking below benchmarks every non-deprecated model we track on a realistic workload shape, recomputed automatically whenever pricing changes.
- Cheapest model for a chatbot, 200 input + 100 output per turn
- Cheapest model for RAG, 4k input + 400 output
- Cheapest model for batch summarization, 8k input + 500 output
- Cheapest model for code review, 3k input + 600 output
- Cheapest model for translation, 1k input + 1k output
- Cheapest model for a coding assistant, 500 input + 1.5k output
- Cheapest long-context model, 50k input + 1k output, 100k+ windows only
- Cheapest reasoning model, 500 input + 3k output
Pricing and cost
- What is the cheapest AI model?, current cheapest models across all providers, updated when pricing changes
- How much does GPT-4o cost per million tokens?, the math broken down
- How much does prompt caching save?, per-provider cached vs standard rates
- Which providers offer batch-API discounts?, the 50%-off async tier explained
Comparisons
Head-to-head pricing breakdowns with cost-at-workload tables showing which model wins at five common workload shapes.
- Claude Opus 4.8 vs GPT-4o
- GPT-4o vs Claude Sonnet
- GPT-4o vs GPT-4o-mini
- Gemini 2.5 Flash vs Pro
- Claude Sonnet 4.5 vs GPT-4o-mini
- GPT-5 vs Claude Opus 4.8
- GPT-5 vs Gemini 3.1 Pro
- DeepSeek V3 vs GPT-4o-mini
- Llama 3.1 405B vs GPT-4o
- o3 vs Claude Opus 4.8
- Gemini 2.5 Pro vs Gemini 3.1 Pro
Reference
- Longest context window in 2026, current leaders ranked
- How to count tokens, programmatic options for each provider
- Methodology, how we count tokens, source pricing, and label confidence
- News and pricing updates, model launches and price changes as they happen
- Pricing changelog, every pricing update logged with date