LLM Pricing Per Million Tokens: 2026 Guide

TLDR

GPT-4.1 Nano is the price floor at $0.10 input and $0.40 output per million tokens. Claude ranges from Haiku through Fable 5, with output carrying the larger share of the bill. Batch processing and prompt caching can cut real bills below listed prices.

Key Takeaways

• GPT-4.1 Nano is the cost floor in this comparison at $0.10 input and $0.40 output per million tokens.
• Claude Sonnet 4.6 is priced at $3/$15 per million tokens, while Fable 5 is priced at $10/$50 per million tokens.
• Output tokens cost 4 to 8 times more than input tokens.
• One million tokens is roughly 750,000 English words, or about 1,500 pages of plain text.
• Batch processing and prompt caching routinely cut real bills 30 to 60% below sticker price.

How Much 1 Million Tokens Costs on Claude

Claude Haiku 4.5 starts at $1/1M input tokens. Related analysis

Claude Sonnet 4.6 is priced at $3/$15 per million tokens. Related analysis

The Claude Opus 4.x line is priced at $5/$25 per million tokens. Related analysis

Fable 5, the top Claude tier, is priced at $10/$50 per million tokens. Related analysis

The practical answer depends on which Claude model you use and whether your workload generates much text. A support workflow that sends long documents into the model but asks for terse outputs will lean toward input cost. A coding agent, research assistant, or drafting workflow can push far more spend into output.

Claude's range also tells you something useful about model selection. Haiku is the cost-conscious choice when speed and a low input price matter most. Sonnet is a middle ground for teams that want stronger performance without paying premium-tier rates. Opus and Fable 5 belong in workflows where the quality of the answer is worth a much larger bill.

You can compare the full Anthropic lineup in the Anthropic API pricing table. If your decision is specifically about Claude models, the Claude pricing guide is the more focused place to start.

The Cheapest and Most Expensive Models Right Now

The cheapest capable model is GPT-4.1 Nano at $0.10 input and $0.40 output per million tokens. Related analysis

At the other end, Fable 5, the top Claude tier, is priced at $10/$50 per million tokens. Related analysis

That is the spread developers are working with. The floor is cheap enough that many high-volume, narrow tasks are economically viable. Classification, routing, extraction, short summaries, and structured transformations all become easier to justify when the model cost stays close to the floor.

Premium models change the math quickly. They are usually chosen for a reason: harder reasoning, better writing, deeper analysis, stronger tool use, or a higher tolerance for ambiguous instructions. But those benefits have to show up in the product. Paying premium rates for a task that a smaller model already handles is a quiet way to turn healthy gross margins into a literal money pit.

The most expensive model is not automatically the wrong choice. It may be the right model for the last step in a workflow, where the system has already narrowed the task and the output goes directly to a customer, analyst, or executive. It is a poor default for every step that comes before it.

Model routing is where this becomes a product decision rather than a procurement exercise. Use a cheaper model to handle routine work. Send only the uncertain, high-value, or customer-visible cases to the premium tier. The provider with the lowest listed rate does not always win. The winning setup is the one that gets the required quality at the lowest total cost.

Representative Models Compared

The table is useful because it removes the language around each product and leaves the commercial decision sitting in front of you.

GPT-4.1 Nano is the clear price-floor option in this representative set. Claude Haiku gives you a low-cost Claude entry point on input. Sonnet occupies the middle of the Claude range, while Opus and Fable 5 are premium choices with correspondingly higher output prices.

Do not treat the input column as the whole bill. Many applications receive a short response and can mostly budget around input. Others create large outputs. Code generation, long-form drafting, agent traces, detailed analysis, and repeated tool calls can all move the output side from a rounding error to the dominant expense.

The model name matters less than your request pattern. Before choosing a default, inspect what your application sends and what it receives. A system that reads a lot and says little needs a different pricing model from one that produces long answers from compact prompts.

For a wider provider view, use the LLM API pricing comparison. It is easier to compare models after you have decided whether your workload is input-heavy, output-heavy, or split between the two.

What 1 Million Tokens Actually Is

One million tokens is roughly 750,000 English words, or about 1,500 pages of plain text. Related analysis

Tokens are the units an LLM reads and generates. They are not words, characters, or pages, although those comparisons help when you are estimating usage before you have production logs.

The words-and-pages equivalence gives you a rough mental model for volume. A million tokens is a large amount of text. Yet API products can consume that amount faster than expected when they include long source documents, repeated instructions, conversation history, retrieved context, tool results, or large generated outputs.

That is why token budgets deserve the same scrutiny as cloud infrastructure budgets. The first version of an application may use a small prompt and return a compact answer. The production version often accumulates context. Product teams add personalization. Developers add system instructions. Customers upload files. The model sees more of the world, and the bill follows.

The fix is rarely "use less AI." It is usually more specific. Remove context that does not affect the answer. Summarize older conversation history. Keep retrieved passages relevant. Put stable instructions in a cached prompt when the provider supports it. Route simple tasks to lower-cost models.

A million tokens is a useful pricing unit because it makes those choices comparable across providers. It is not a promise about your monthly bill. Your bill comes from the token shape of your product.

Complete Pricing Table: All Major LLM APIs

This table covers six providers and 20+ models. All prices are per 1 million tokens. "Input" is what you send to the model (prompts, context, documents). "Output" is what the model generates back. Where a provider offers tiered pricing based on context length, both tiers are listed.

OpenAI

Anthropic

Google

Meta (Llama 4, via inference providers)

Mistral

Cohere

Cheapest Models by Price Tier

Why Output Tokens Cost 4-8x More Than Input

Every LLM provider charges more for output tokens than input tokens. GPT-4.1 charges $2 for input but $8 for output. Opus 4.6 charges $5 for input but $25 for output. This isn't arbitrary. It reflects how the models work.

When you send input tokens (your prompt), the model processes them all at once in parallel. The GPU crunches through your entire prompt in a single forward pass. It's fast and computationally efficient.

Output tokens are different. The model generates them one at a time, sequentially. Each new token depends on every token that came before it. The model can't skip ahead or parallelize this step. Token 50 requires tokens 1 through 49 to exist first. This sequential dependency means each output token requires a separate forward pass through the model, using far more compute per token than input processing.

This has a direct impact on your bill. A typical API call with 2,000 input tokens and 500 output tokens on GPT-4.1 costs $0.004 for input and $0.004 for output. Equal cost despite 4x fewer output tokens. If your responses average 1,000 output tokens, output is 67% of your total cost.

The practical takeaway: controlling output length is the single most effective way to reduce costs. Tell the model to be concise. Use structured output (JSON) instead of prose. For classification, request a single label instead of an explanation. Set max_tokens to prevent runaway responses. A 10-class classifier using structured output typically uses 5-20 output tokens instead of 50-200 with free-form text. At scale, that's a 10x cost reduction on your most expensive token type.

How to Calculate Your Monthly Cost: 3 Real Scenarios

Token pricing means nothing in isolation. Here's what three common workloads cost per month across different models, with the math shown.

Batch and Caching Discounts That Change the Math

The per-million-token prices above are sticker prices. Three discount mechanisms can cut your actual costs by 50-90%. Ignoring them means overpaying.

OpenAI Batch API: 50% off everything. Submit requests as a JSONL file and get results within 24 hours. Every token costs half. GPT-4.1 drops from $2/$8 to $1/$4. GPT-4.1 Nano drops to $0.05/$0.20. Use this for any workload that doesn't need real-time responses: data extraction, content generation, nightly processing, bulk classification. It's the single biggest cost reduction available from any provider.

Prompt caching: 50-90% off input tokens. Both OpenAI and Anthropic cache repeated prompt prefixes automatically. If your system prompt, few-shot examples, or document context stays the same across requests, subsequent calls pay a reduced rate on those cached tokens. OpenAI's discounts vary by model: GPT-5 gets 90% off cached reads, GPT-4.1 gets 75% off, o-series gets 50% off. Anthropic offers 90% off cached input tokens across all Claude models. Structure your prompts with stable content first and variable content last to maximize cache hits.

Google's free tier. Gemini 2.5 Flash is free under certain rate limits (roughly 15 RPM and 1M TPM). For prototyping, low-volume internal tools, or development environments, this is zero-cost access to a model that competes with GPT-4.1 Nano in quality. No credit card required.

These discounts stack in some cases. A cached batch request through OpenAI's GPT-4.1 pays $0.25 per million cached input tokens and $4 per million output tokens. That's 87.5% off standard input pricing. The sticker prices at the top of this page are the ceiling, not the floor.

Which Provider Is Cheapest for Your Use Case?

Stop comparing models in isolation. Match the model to your task. Here are clear winners in each category.

Input Output Tokens: What You Pay For and Why They Differ

Every API bill splits into two line items: input tokens and output tokens. Input tokens are everything you send to the model. That includes your system prompt, the conversation history, retrieved documents, few-shot examples, and the user's current message. Output tokens are only the text the model writes back. The split matters because the two are priced differently and behave differently at scale.

A quick worked example. You send GPT-4.1 a 3,000-token prompt and it returns a 700-token answer. Input cost is 0.003 x $2 = $0.006. Output cost is 0.0007 x $8 = $0.0056. Almost a 50/50 split even though the output is four times shorter, because output is priced four times higher. Knowing which side dominates your workload tells you where to optimize. Long-context RAG is input-heavy. Chatty assistants and code generation are output-heavy.

Here is the input output token spread for the models most teams actually deploy, per the maintained table above:

How Much Is 1 Million Tokens (and 2 Million, and 10 Million)?

In plain dollars, the answer depends entirely on the model and whether the tokens are input or output. One million tokens is approximately 750,000 words. On GPT-4.1 Nano, a million input tokens costs ten cents and a million output tokens costs forty cents. On Opus 4.6, the same million input tokens costs five dollars and a million output tokens costs twenty-five dollars. The table below shows the all-in cost for a few round numbers, assuming a typical 80% input / 20% output mix.

The cost per token, then, ranges from about $0.0000001 (GPT-4.1 Nano input) to $0.000025 (Opus 4.6 output). Those numbers are tiny per token, which is exactly why providers quote per million. Confirm the current rate on the provider's own pricing page before you budget, since model lineups and prices shift through the year. See OpenAI API pricing and Anthropic API pricing for the per-provider breakdowns we keep updated.

How to Compare LLM API Pricing Without Fooling Yourself

Start by putting every candidate model into the same unit: cost per million input tokens and cost per million output tokens. That is the normalization layer. Without it, product pages can make similar rates look unrelated.

Then inspect your own workload. What does a typical request contain? How much reusable context is included? How long is the output? Does the workflow call the model once, or does it make repeated calls behind the scenes? Which requests need the highest-quality result?

You do not need perfect production data to make an initial decision. You need an honest request shape. Build from the type of text your product will send and receive, then choose a default model that fits the common case.

Keep the premium option available. The cheapest model can lose when it creates enough errors, retries, or manual review to erase its rate advantage. The premium option can lose when it handles routine work that did not need it. Good model selection is a portfolio decision.

It also changes over time. Product usage evolves. Providers update models. A workflow that starts with simple prompts can grow into a system with retrieval, file uploads, multiple tools, and customer-specific context. Revisit the pricing table when the product changes, rather than only when finance asks why spend moved.

The LLM token pricing guide covers the mechanics behind token billing. Use it alongside this table when you need to turn model rates into a product-level estimate.

Related Resources

Sources

• OpenAI API pricing (official)
• Anthropic Claude pricing (official)
• Google Gemini API pricing (official)
• PE Collective LLM pricing table: $0.10 input and $0.40 output
• PE Collective Anthropic API pricing: $3/$15 per million tokens
• PE Collective LLM pricing table: 30 to 60% below sticker price