LLM API price comparison — find the cheapest model, no upload
Pick the cheapest AI model for today's work. Prices are per million tokens, in USD. Click a tier to filter, or tap the live link to check real-time pricing on official pages.
How to read these prices
Input is what you pay to send a prompt in. Output is what you pay for the model's reply. Output is almost always more expensive, sometimes 3 to 10 times the input price.
The blended price folds both into one number using a 3:1 input-to-output mix, which is close to how most real apps behave.
All numbers are a reference snapshot from public pricing pages. Prices change often, so always confirm on the official page before you commit budget.
Why the cheapest model is usually enough
Most everyday tasks, like classification, summarization, drafting, and simple code, run fine on a flash or mini tier model. Reserve the premium models for hard reasoning or long, nuanced output. Routing 80% of requests to a cheap model and 20% to a flagship one is the fastest way to cut your AI bill without hurting quality.
How the blended price is calculated
Different models charge different rates for input and output tokens, and output is usually several times more expensive. To make models comparable on one number, this tool uses a blended price: 75% of the input rate plus 25% of the output rate, per million tokens — a rough 3:1 input-to-output ratio typical of chat and coding workloads. If your workload is output-heavy, weight the output rate more heavily when you compare.
How to use the tiers
- Cheapest — the lowest blended prices; ideal for drafts, classification, and high-volume routine tasks.
- Middle — the sweet spot for everyday assistant work.
- Priciest — frontier models for hard reasoning; use them where quality, not volume, is the bottleneck.
- Save tokens — shows how much more work the same budget buys at the cheaper tier.
Prices are a dated snapshot for orientation, not a live feed — each card links to the provider's official pricing page, and the header links to OpenRouter for real-time aggregation. Always confirm on the official page before committing to a model.
FAQ
Why do prices differ so much between models?
Frontier models cost more because of larger compute per token; smaller or distilled models are cheap because they do less work per request. Matching the model to the task difficulty is where the savings are.
Do free tiers really cost nothing?
Some providers offer genuinely free flash-tier models with rate limits. Check the linked official page for current limits.
Is the token counter included?
No, this is a price comparison tool. Use your provider's tokenizer to count tokens for a specific prompt.
Can I use this for chargeback or budget approval?
Use it as a fast sanity check, not as an invoice. Prices are transcribed from public pricing pages and can lag a provider's announcement by a day or two; enterprise discounts, batch pricing and cached-input rebates are not modelled. Always confirm against the provider's own billing page before you commit to a number in writing.
Which models are covered?
The table covers the major hosted frontier and mid-tier models plus popular open-weight options you can self-host. Input and output prices are listed separately because output tokens are usually three to five times more expensive, and that is where most budget surprises come from.
People also use this for…
Every tool below is also 100%% local: no upload, no account, no server copy — the same privacy guarantee you get on this page.
- Count words before you count tokens
A rough rule: 1 token ≈ 0.75 English words. Get the exact count first. - Clean up the API response you are parsing
Pretty-print and validate the JSON your model returns. - Rotate the API key password
Generate a fresh credential with the browser's crypto API.
How to use the LLM price picker
- Enter your monthly budget in USD.
- Pick a strategy: cheapest (lowest blended price), mid-range (balanced), most expensive (premium models), or save tokens (what you would save versus the mid-range option).
- Read what your budget buys at that price point, expressed in tokens per month and the equivalent number of typical conversations.
- Adjust the budget to see where the breakpoint sits — the point where a better model stops being affordable is usually more useful than the headline price.
How the numbers are calculated
Each model in the table carries an input price and an output price per million tokens. The tool blends them on a configurable ratio of input to output tokens, because in real workloads output tokens usually cost more and dominate the bill. Your budget divided by the blended price gives the token volume you can afford, and that volume is converted into an approximate number of exchanges using a typical message length. Everything is arithmetic in your browser — no API call, no account, no key.
Read this before relying on the numbers
Prices on this page are a snapshot, not a live feed. Model pricing changes often — new tiers appear, context windows are repriced, and providers run promotions. Treat the output as a way to compare orders of magnitude and to sanity-check a plan, and always confirm the current figure on the provider's own pricing page before committing a budget. The tool is also deliberately simple: it does not model batch discounts, cached-token pricing, context-window minimums, or per-request overheads, all of which can move a real invoice by a wide margin.
What actually drives LLM cost
- Output tokens usually cost three to five times input tokens, so verbose answers are expensive.
- Context size — a long system prompt or a chat history you resend every turn multiplies input tokens fast.
- Retries and agent loops quietly double or triple usage.
- Model choice can span two orders of magnitude for tasks where the cheapest model is perfectly adequate.
- Caching, where available, is usually the single biggest saving for repeated prompts.
FAQ
How do I estimate my token count?
A rough rule for English: one token is about four characters, or three quarters of a word. A 1,000-word document is roughly 1,300–1,500 tokens. Non-English text and code tokenise less efficiently.
Why is the output price higher than the input price?
Generating tokens requires a forward pass per token, while reading the prompt is processed in parallel — providers price the two separately to reflect that cost difference.
Does it include free tiers or open-weight models?
The comparison focuses on hosted API pricing. Self-hosted open-weight models have a completely different cost structure — GPU time rather than per-token billing — and are not directly comparable.
Is my budget figure uploaded anywhere?
No. The calculation runs locally and the number you type is never sent anywhere.