We use cookies for analytics and personalized ads. We use cookies for basic analytics and, where we show ads, to personalize them. Privacy PolicyRead our Privacy Policy.
Paste your text to see its token count and what it costs on each major model. Nothing leaves your browser.
Presets fill in realistic example text plus a matching reply length and monthly volume, so the numbers below aren't a guess from zero.
Output tokens are billed at 4-6x the input rate on most models — this number moves cost more than the prompt itself.
Turns a single prompt's cost into a realistic monthly bill for budgeting.
The link only carries your output-token, volume, currency and model settings — never the text you pasted.
(input tokens ÷ 1M × input rate) + (output tokens ÷ 1M × output rate), then × your monthly requests.
Switching to Gemini 2.5 Flash-Lite is roughly 98% cheaper at this volume ($5.28 → $0.1060/mo) — worth it if the task doesn't need GPT-5.6 Sol's extra reasoning quality.
Every LLM API bills and limits by tokens, not characters or words.
Check whether a document or chat history fits inside a model's limit.
Turn a token count into a dollar estimate before you send it.
The same prompt can cost 10x more or less by provider.
Spot instructions quietly eating your budget.
Token usage shows up differently depending on the AI tool you're in.
Every request and reply is billed by input and output tokens separately. Long system prompts and chat history stack up fast.
Large document and codebase uploads are where token counts spike — cached reads make repeat context cheaper.
Multimodal input (images, video, long PDFs) gets converted into tokens too, so one attachment can dominate a request's cost.
Token usage comes from agent actions reading your codebase — a multi-file refactor can burn more tokens than the prompt suggests.
Off-peak and peak pricing can differ by roughly 2x on the same model, so token usage that looks cheap at 3am can cost noticeably more during business hours.
Any prompt, code block, JSON payload, or transcript works.
Tokens, words, and characters update as you type — nothing leaves your browser.
Turns a single prompt's cost into a real monthly bill estimate.
Sorted and highlighted so the cheapest and priciest models jump out immediately.
Use this quick counter for a fast check, then jump into a provider guide below for pricing tiers, caching rules, and cost math specific to that API.
Compare GPT-5.6 models and estimate OpenAI API cost by request and month.
Choose among Claude models and budget long context, caching, and output.
Compare Gemini models and plan text, multimodal, caching, and service-tier cost.
Estimate OpenAI, Claude, Gemini, and ChatGPT-style API cost from token counts.
Convert token budgets into rough word counts, including 1 million token to words.
Rates per 1M tokens. Estimated monthly cost updates live against your text, output length, and volume above.
Teal row = cheapest, orange row = most expensive within the selected tab, at your current volume.
Bills at the routed model's own rate above, plus a $0.25/1M token platform fee on Teams & Enterprise.
| Model | Provider | Context / Max out | Input | Cache read | Output | Est. monthly |
|---|---|---|---|---|---|---|
| GPT-6 Astra Newest flagship, rolling out. Long context after 272K input: input $20, cache read $2, output $75 | OpenAI | 1,050K / 128K | $10.000 | $1.000 | $50.00 | $13.20 |
| GPT-5.6 Sol Promotional rate cut ~20% on 21 Aug 2026, guaranteed at least through 21 Nov 2026 (list price is $5/$30) | OpenAI | 1,050K / 128K | $4.000 | $0.400 | $20.00 | $5.28 |
| GPT-5.6 Terra Balanced mid-tier option | OpenAI | 1,050K / 128K | $2.000 | $0.200 | $12.00 | $3.16 |
| GPT-5.6 Luna Lightweight, high-volume tasks | OpenAI | 1,050K / 128K | $0.200 | $0.020 | $1.20 | $0.3159 |
| GPT-5.6 Cyber Security & enterprise-focused variant | OpenAI | 1,050K / 128K | $12.500 | $1.250 | $75.00 | $19.74 |
| GPT-5.5 Previous flagship generation | OpenAI | 1,050K / 128K | $5.000 | $0.500 | $30.00 | $7.90 |
| GPT-5.5 Pro Extended reasoning, premium tier | OpenAI | 1,050K / 128K | $30.000 | Not offered | $180.00 | $47.39 |
| GPT-5.4 Prior-generation standard model | OpenAI | 1,050K / 128K | $2.500 | $0.250 | $15.00 | $3.95 |
| GPT-5.4 Mini Compact, cost-efficient | OpenAI | 400K / 128K | $0.750 | $0.075 | $4.50 | $1.18 |
| GPT-5.4 Nano Smallest, cheapest OpenAI tier | OpenAI | 400K / 128K | $0.200 | $0.020 | $1.25 | $0.3289 |
| GPT-5.4 Pro Prior-gen premium reasoning | OpenAI | 1,050K / 128K | $30.000 | Not offered | $180.00 | $47.39 |
| GPT-5-Codex Built for agentic coding tasks | OpenAI | 1,050K / 128K | $1.250 | $0.130 | $10.00 | $2.62 |
| GPT-5 Pro Deep reasoning workloads | OpenAI | 1,050K / 128K | $15.000 | Not offered | $120.00 | $31.49 |
| Claude Fable 5.1 Current generation; cache reads 75% cheaper than Fable 5 (0.025x input, vs the standard 0.1x) | Anthropic | 1M / 128K | $10.000 | $0.250 | $50.00 | $13.20 |
| Claude Mythos 5.1 Same model as Fable 5.1 with relaxed cyber/biosafety limits — invite-only | Anthropic | 1M / 128K | $10.000 | $0.250 | $50.00 | $13.20 |
| Claude Fable 5 Prior generation, superseded by Fable 5.1 | Anthropic | 1M / 128K | $10.000 | $1.000 | $50.00 | $13.20 |
| Claude Mythos 5 Prior generation, superseded by Mythos 5.1 | Anthropic | 1M / 128K | $10.000 | $1.000 | $50.00 | $13.20 |
| Claude Sonnet 5 Balanced flagship | Anthropic | 1M / 128K | $2.000 | $0.200 | $10.00 | $2.64 |
| Claude Opus 5 High-capability analysis | Anthropic | 1M / 128K | $5.000 | $0.500 | $25.00 | $6.60 |
| Claude Opus 4.8 Enterprise-grade workflows | Anthropic | 1M / 128K | $5.000 | $0.500 | $25.00 | $6.60 |
| Claude Sonnet 4.6 Prior-gen balanced model | Anthropic | 1M / 128K | $3.000 | $0.300 | $15.00 | $3.96 |
| Claude Haiku 4.5 Fast, lightweight tasks | Anthropic | 200K / 64K | $1.000 | $0.100 | $5.00 | $1.32 |
| Gemini 3.8 Flash Newest fast-tier model. Introductory rate through 31 Dec 2026, then doubles to $1.50/$7.50 | 1,048K / 65K | $0.750 | $0.075 | $3.75 | $0.9897 | |
| Gemini 3.7 Flash Introductory rate through 31 Dec 2026, then doubles to $1.50/$7.50 | 1,048K / 65K | $0.750 | $0.075 | $3.75 | $0.9897 | |
| Gemini 3.6 Flash Introductory rate through 31 Dec 2026, then doubles to $1.50/$7.50 | 1,048K / 65K | $0.750 | $0.075 | $3.75 | $0.9897 | |
| Gemini 3.5 Flash Multimodal grounding support | 1,048K / 65K | $1.500 | $0.150 | $9.00 | $2.37 | |
| Gemini 3.5 Flash-Lite Built for high-volume jobs | 1,048K / 65K | $0.300 | $0.030 | $2.50 | $0.6559 | |
| Gemini 3.1 Pro Preview Preview-tier reasoning model | 1,048K / 65K | $2.000 | $0.200 | $12.00 | $3.16 | |
| Gemini 3.1 Flash-Lite Cost-efficient multimodal | 1,048K / 65K | $0.250 | $0.025 | $1.50 | $0.3949 | |
| Gemini 2.5 Pro Prior-gen flagship reasoning | 1,048K / 65K | $1.250 | $0.125 | $10.00 | $2.62 | |
| Gemini 2.5 Flash Configurable thinking budget | 1,048K / 65K | $0.300 | $0.030 | $2.50 | $0.6559 | |
| Gemini 2.5 Flash-Lite Lowest-cost Gemini tier | 1,048K / 65K | $0.100 | $0.010 | $0.40 | $0.1060 | |
| DeepSeek V4 Flash Off-peak rate — peak ≈2x higher. Thinking/non-thinking modes, tool calls | DeepSeek | 1M / 384K | $0.220 | $0.007 | $0.66 | $0.1759 |
| DeepSeek V4 Pro Off-peak rate — peak ≈2x higher. Higher-precision reasoning | DeepSeek | 1M / 384K | $0.660 | $0.022 | $1.98 | $0.5277 |
| DeepSeek V4 Flash Vision (exp) Off-peak rate — peak ≈2x higher. Experimental vision-token support | DeepSeek | 1M / 384K | $0.220 | $0.007 | $0.66 | $0.1759 |
| Grok 4.7 Current flagship. Long context ≥200K input tokens: input $4, cache $1, output $12 | xAI | 500K | $2.000 | $0.500 | $6.00 | $1.60 |
| Grok 4.3 Mid-tier, largest context window in the lineup. Long context ≥200K input tokens: input $2.50, cache $0.40, output $5 | xAI | 1M | $1.250 | $0.200 | $2.50 | $0.6745 |
| Grok Build 0.1 Coding-focused, cheapest xAI tier. Long context ≥200K input tokens: input $2, cache $0.40, output $4 | xAI | 256K | $1.000 | $0.200 | $2.00 | $0.5396 |
| Sonar Base web-search model. Excludes the $5-$12 per 1K request search fee | Perplexity | 127K | $1.000 | Not offered | $1.00 | $0.2796 |
| Sonar Reasoning Pro Adds reasoning before answering. Excludes the $6-$14 per 1K request search fee | Perplexity | 128K | $2.000 | Not offered | $8.00 | $2.12 |
| Sonar Pro Highest-quality search answers. Excludes the $6-$14 per 1K request search fee | Perplexity | 200K | $3.000 | Not offered | $15.00 | $3.96 |
| Cursor (Auto / Composer) | Bills at the routed model's own rate above, plus a $0.25/1M token platform fee on Teams & Enterprise. | — | ||||
All providers — Input, cache read, and output rates per 1M tokens.
Four pieces typically make up what you're billed for on any given call. Knowing which one is growing tells you what to trim first.
The fixed instructions sent with every call — tone, rules, output format. It's billed on every single request even though it never changes.
e.g. "You are a support agent. Reply in under 100 words, always offer a refund or reship."
Trap: letting the system prompt grow unchecked over months of tweaks — it's the easiest place to save cost with caching or trimming.
The part that actually changes per call — the question, the document, the ticket text. This is what most people mean when they say "my prompt."
e.g. the customer's message, a pasted contract, a code diff.
Trap: pasting an entire document when only one section is relevant — every extra paragraph is billed.
Extra chunks of text pulled in from a knowledge base or search step and stuffed into the prompt before it reaches the model.
e.g. 5 document snippets retrieved for a Q&A bot.
Trap: retrieving more chunks than the model needs "just in case" — often the largest, least-controlled cost driver.
What the model generates back. It's billed separately from input, usually at 4-6x the input rate on the same model.
e.g. a 400-word drafted reply, a JSON object, a summary.
Trap: not capping max output length — a model that rambles costs more than one instructed to be concise.
Say each ticket sends roughly 350 input tokens (a short system prompt plus the customer's message) and gets back 220 output tokens. On Claude Haiku 4.5 that's (350 ÷ 1e6 × $1) + (220 ÷ 1e6 × $5) ≈ $0.00135 per ticket. Multiplied by 12,000 tickets, that's roughly $16.20/month — small enough that the team could safely upgrade to a stronger model for the ~5% of tickets that get escalated, without materially changing the bill. This is exactly the trade-off the recommendation card above is built to surface for your own numbers.
Cache-read pricing, context window vs. max output, long-context tiers, and reasoning-effort cost — plus a model-by-model comparison table — live on the dedicated guide for each provider so they can go deeper than a homepage summary can.
Replit doesn't bill by tokens — it charges a fixed rate per action ("checkpoint"), win or fail.
Scaffolding touches many files — expect a high Agent checkpoint count.
Failed attempts still bill — five retries cost five checkpoints.
Small single-file changes usually route through cheaper Assistant checkpoints.
Multi-file Agent work is where monthly credit pools disappear fastest.
Add a live token counter to your own blog or app in one line of code — free, no signup, no branding removed.
<iframe src="https://apextechjournal.com/widget/token-counter.html" width="100%" height="280" style="border:none; border-radius:10px;"></iframe>