OpenAI API Pricing vs Claude and Gemini: Costs Compared

OpenAI API pricing, like Claude and Gemini API pricing, is a per-token charge: you pay for every million tokens you send to a model and every million tokens it writes back. This guide compares the current models from OpenAI, Anthropic and Google side by side and explains the discounts, extra fees, free tiers and spending caps. It also works out monthly costs for real kanban board workloads.

Prices checked on 20 October 2026 on each provider’s official pricing page. Providers change prices and retire models often, and Google has already announced price rises for several Gemini models, so check the linked pages before you budget.

OpenAI, Claude and Gemini API pricing compared

All prices are standard (non-batch) rates in US dollars per million tokens. “Cached input” is the price for input the provider has already cached. For Claude it is the cache hit price.

Sources: OpenAI API pricing, Claude API pricing and Gemini API pricing.

Google’s page lists scheduled increases. Gemini 4 Argon rises to $4 input and $20 output on December 1, 2026. Gemini 3.8 Flash and Gemini 3.6 Flash rise to $1.50 input and $7.50 output on January 1, 2027.

How LLM API pricing works

Every request has input tokens and output tokens. A token is a piece of a word, and the provider counts both sides of each request:

• Input tokens are everything you send: the system prompt, your instructions, the task text and any earlier messages.
• Output tokens are everything the model writes. They cost four to about eight times more per token than input on the models above.
• Reasoning or thinking tokens count as output. OpenAI’s reasoning guide says hidden reasoning tokens are billed as output tokens, and Google’s price list says its output price includes thinking tokens.
• Cached input is a repeated prompt prefix the provider has stored. Reading cached input costs a fraction of the normal input price. Writing input to the cache can cost more than the normal price.

The cost of one request is:

cost = (input tokens / 1,000,000 × input price) + (output tokens / 1,000,000 × output price)

Each provider splits text into tokens in its own way. Anthropic’s pricing page says Claude 4.7 and later models, which include every Claude 5 model above, produce about 30% more tokens for the same text than older Claude models. A Claude model and an OpenAI model with the same price per token can therefore cost different amounts for the same task.

OpenAI API pricing for GPT-6 models

OpenAI’s model list describes three flagship models. GPT-6 Astra is OpenAI’s most capable model, GPT-6.1 Sol gives near-Astra performance at a lower cost, and GPT-6 Luna is the most efficient model for focused, high-volume tasks. All three have a 1.05 million token context window and up to 128,000 output tokens.

OpenAI sells the same models at several processing tiers:

• Batch and Flex cost 50% of standard rates. gpt-6-luna drops to $0.05 input and $0.25 output per million tokens.
• Fast costs about twice the standard rate. gpt-6-luna costs $0.20 input and $1.00 output.
• Ultrafast costs six times the standard rate on gpt-6-astra, at $60 input and $300 output.

The gpt-6-luna model page sets out the caching and long-prompt rules. Cached input costs 10% of the normal input rate, and cache writes cost 1.25 times the normal input rate. If a prompt has more than 272,000 input tokens, the whole request is billed at twice the input and cache rates and 1.5 times the output rate.

Older OpenAI models are still on the price list. The GPT-5.6 models cost $2 input and $12 output for gpt-5.6-terra and $0.20 and $1.20 for gpt-5.6-luna. gpt-5.6-sol has a promotional price of $4 and $20, which OpenAI says lasts at least through November 21, 2026. gpt-5-nano costs $0.05 input and $0.40 output, and gpt-5.4-mini costs $0.75 and $4.50.

Claude API pricing (Anthropic)

Anthropic’s current lineup runs from Claude Haiku 5.5 at the low end to Claude Fable 5.1 at the top. Anthropic also lists Claude Mythos 5.1 at the same price as Fable 5.1, but only verified organizations can use it.

Claude Haiku 5.5 has two price levels. Prompts up to 100,000 tokens cost $0.10 input and $0.50 output per million tokens. A request whose prompt is over 100,000 tokens costs $0.50 input and $2.50 output. Cached tokens count toward the 100,000 limit. Other Claude models from 4.6 onward charge the same rate across the full 1 million token context window.

Claude API cost changes with these options:

• Batch API: 50% off input and output. Claude Sonnet 5.5 drops to $1 input and $5 output.
• Prompt caching: a 5-minute cache write costs 1.25 times the base input price, and a 1-hour write costs 2 times. A cache hit costs 0.1 times base input on most models, 0.05 times on Opus 5.5 and Sonnet 5.5, and 0.025 times on Fable 5.1.
• Fast mode (faster output at a higher price): Claude Opus 5.5 costs $8 input and $40 output. You cannot use fast mode with the Batch API.
• US-only inference: setting inference_geo to us multiplies all token prices by 1.1 on Claude 4.6 and later models.

Older Claude models remain on sale at higher prices. Claude Sonnet 4.6 costs $3 input and $15 output, and Claude Haiku 4.5 costs $1 and $5.

Gemini API pricing

Google’s Gemini price list has a free tier and four paid options: Standard, Batch, Flex and Priority.

• Free tier: available for Gemini 3.8 Flash, Gemini 3.6 Flash, the Flash-Lite models, Gemini 2.5 Flash and Gemini 2.5 Pro, but not for Gemini 4 Argon. Google says it uses free-tier content to improve its products and does not use paid-tier content.
• Batch: 50% of standard for most text models. Gemini 3.6 Flash costs $0.375 input and $1.875 output on Batch until December 31, 2026.
• Flex: the same rates as Batch for most models.
• Priority: above standard. Gemini 3.6 Flash costs $1.35 input and $6.75 output.

Gemini context caching has two charges: a per-token price for cached input and an hourly storage price. On Gemini 3.6 Flash the storage price is $0.50 per million tokens per hour until the end of 2026. Gemini 3.1 Pro Preview also has a long-prompt rate: $2 input and $12 output up to 200,000 tokens, and $4 input and $18 output above that.

Extra fees: web search, tools and data residency

Token prices are not the whole bill when a model uses built-in tools:

Anthropic does not charge for code execution when the same request also uses its web search or web fetch tool. Giving a model tools also adds input tokens to each request. On Claude Haiku 5.5, Anthropic adds a hidden tool-use prompt of at least 286 tokens.

Embedding, image and speech-to-text prices

OpenAI and Google also sell models for text embeddings, image generation and transcription. Anthropic’s price list has no embedding, image generation or speech-to-text models.

OpenAI prices images by token, and its image generation guide has a calculator for the cost per image.

Monthly cost examples for a kanban board

These examples use the standard prices and the cost formula above. The token counts are estimates for typical tasks. Your own counts depend on how long your tasks and comments are, and on how many reasoning tokens a model uses.

Example 1: a task-summary assistant

A team of 20 people each asks for 10 task summaries per working day, over 22 working days. Each request sends about 1,500 input tokens (instructions, task details and comments) and gets back about 300 output tokens.

One request costs $0.0003 on gpt-6-luna or Claude Haiku 5.5, and $0.006 on gpt-6.1-sol or Claude Sonnet 5.5. For the month:

• Requests: 20 × 10 × 22 = 4,400 per month
• Input: 4,400 × 1,500 = 6,600,000 tokens (6.6 million)
• Output: 4,400 × 300 = 1,320,000 tokens (1.32 million)

The same workload costs 100 times more on gpt-6-astra or Claude Fable 5.1 than on gpt-6-luna or Claude Haiku 5.5.

Example 2: daily standup summaries for 30 boards

A daily summary for each of 30 boards, every working day, means 30 × 22 = 660 requests a month. Each summary reads about 8,000 input tokens of board activity and writes about 800 output tokens.

Over a month, the summaries use 5.28 million input tokens and 0.528 million output tokens, which gives these totals:

Input is about two-thirds of the cost on each model. Output tokens cost five times more than input on these models, but a summary reads about 10 times more tokens than it writes, so input costs about twice as much as output.

Example 3: blog post drafts

An editorial team asks for 40 post drafts a month. Each request sends about 2,000 input tokens and gets back about 3,000 output tokens, so the team uses 0.08 million input tokens and 0.12 million output tokens a month. The monthly totals are about $0.07 on gpt-6-luna, $1.36 on Claude Sonnet 5.5, $2.72 on Claude Opus 5.5 and $6.80 on Claude Fable 5.1. Output is about 88% of the cost on each of these models.

The cheapest LLM API options

Ranked by standard price per million tokens on October 20, 2026, the lowest-cost text models on the three price lists are:

1. gpt-5-nano: $0.05 input, $0.40 output. OpenAI still lists this older model.
2. Gemini 2.5 Flash-Lite: $0.10 input, $0.40 output. Google has not announced a shutdown date for this model.
3. gpt-6-luna: $0.10 input, $0.50 output.
4. Claude Haiku 5.5: $0.10 input, $0.50 output, for prompts up to 100,000 tokens.
5. Gemini 3.1 Flash-Lite: $0.25 input, $1.50 output.

Batch or Flex processing halves the prices of these five models, when you can wait for the result. The Gemini free tier costs nothing, within Google’s rate limits and its terms on product improvement.

Free tiers, trial credits and subscriptions

The three providers offer these free options:

• OpenAI: the API price list has no free token allowance. The only free model on it is omni-moderation-latest.
• Anthropic: new users get a small amount of free credit to test the API.
• Google: new accounts start on the Gemini free tier. Google’s billing page says that billing accounts opened after March 2, 2026 cannot use the $300 Google Cloud welcome credit for the Gemini API.

A chat subscription is a separate product from the API. OpenAI’s billing help page says API usage is billed separately from a ChatGPT subscription, such as Plus or Pro.

On Anthropic’s plans page, Claude Pro costs $20 a month, or $17 a month on annual billing, with no API credits listed. Claude Max starts at $100 a month and includes $100 to $200 a month in API credits for the Claude Platform. Claude Team plans include up to $500 a month in API credits shared by the team.

Usage tiers and spending caps

All three providers place each account on a usage tier. The tier sets the rate limits and the most you can spend in a month:

Sources: OpenAI rate limits, Claude rate limits and Gemini rate limits.

You can also set your own limit below the tier cap:

• OpenAI offers a spend alert, which only sends a notice, and a hard spend limit, which makes requests fail with a 429 error.
• Anthropic lets you set a spend limit below your tier’s cap on the Billing page of the Claude Console.
• Google has experimental project spend caps in AI Studio. Spend can exceed a project cap for about 10 minutes before the cap takes effect. If you prepay, every API key on that billing account stops working when the prepaid balance reaches $0.

How to keep API costs down

• Use a low-cost model for routine jobs. In Example 1 the most expensive models cost 100 times more than the cheapest.
• Cap output length. A maximum token setting limits output tokens, which cost the most per token.
• Send only the text the task needs. Long comment threads and full board histories add input tokens to every request.
• Use Batch or Flex for work that can wait. Overnight reports and bulk tagging cost half as much.
• Start requests with the same fixed text. The provider can then cache that text at 10% of the normal input price or less.
• Watch the long-prompt limits. Claude Haiku 5.5 costs five times more above 100,000 prompt tokens, and GPT-6 models cost more above 272,000.
• Lower the reasoning effort on reasoning models. gpt-6-luna defaults to medium reasoning effort, and its reasoning tokens are billed as output.

Using these models in KanbanWP

KanbanWP’s AI features use your own API key, so you pay the provider’s prices above with no markup. Under KanbanWP > Settings > AI you choose OpenAI, Anthropic, Google Gemini, or Custom for any service that accepts OpenAI-style chat completion requests. If you leave the model field empty, the plugin uses gpt-6-luna for OpenAI, claude-haiku-5-5 for Anthropic and gemini-3.6-flash for Gemini. You can enter another model name instead, such as gemini-2.5-flash-lite.

The plugin sends standard requests to the provider’s chat endpoint:

• It caps each reply at 2,000 output tokens by default, 1,500 for a post outline and 3,000 for a post draft.
• It does not use batch processing or set prompt caching options, so the standard prices apply.
• It does not set a reasoning effort, so OpenAI reasoning models use their default reasoning effort.

To control cost, KanbanWP has its own monthly token limit for each user, separate from the provider’s spending caps. The default limit is 1,000,000 tokens per user per month, and setting it to 0 removes the limit. The plugin’s usage log records the input and output tokens of each request, so you can multiply them by the prices in this guide.

To set up a key, read how to get and use an API key. The settings fields and the connection test are covered in AI providers and API keys. Monthly token limits and the usage log are explained in usage and limits.