LLM API Pricing Comparison: The Honest 2026 Table for 6 Providers

LLM API pricing comparison, short version: for the same 100 million input and 20 million output tokens a month, list prices run from about $20 (GPT-6 Luna, Claude Haiku 5.5) to $2,000 (GPT-6 Astra, Claude Fable 5.1). The mid tier clusters at $2 input and $10 output per million tokens: GPT-6.1 Sol, Claude Sonnet 5.5, Kimi K3.1 Pro and, until November 30, Gemini 4 Argon. Grok 4.7 is cheaper on output. DeepSeek V4 Pro costs about half as much as that group at peak, and half again off-peak. I checked every number on the six official pricing pages on October 27, 2026 (Beijing time). Each one links to its source.

The short answer

• Cheapest per token: GPT-6 Luna and Claude Haiku 5.5, both $0.10 in and $0.50 out.
• Cheapest strong model: DeepSeek V4 Pro at $1.32 in and $3.96 out at peak, half that off-peak. DeepSeek Flash is $0.30 and $1.20. Source: DeepSeek pricing.
• Mid tier: $2 in and $10 out is now the common price for a strong general model.
• Top tier: GPT-6 Astra and Claude Fable 5.1 both cost $10 in and $50 out.
• Watch the dates: three prices on this page are promotions with end dates. They are listed below.

Flagship models side by side

The table above uses the standard rate for short prompts. OpenAI lists GPT-6 Astra at $10 input, $1 cached input and $50 output. Anthropic’s pricing page puts Claude Fable 5.1 at the same $10 and $50. Mythos 5.1 is also $10 and $50, but it has limited availability. Claude Opus 5.5 is much cheaper at $4 and $20.

Google’s Gemini 4 Argon is $2 and $10 through November 30, 2026. From December 1 it goes to $4 and $20. xAI lists Grok 4.7 at $2 in and $6 out with a 500K context window. Moonshot’s Kimi K3.1 Pro is $2 and $10. DeepSeek V4 Pro is $1.32 and $3.96 at peak.

“Flagship” is a loose word here. Opus 5.5, Argon and Grok 4.7 are not in the same price class as Astra and Fable. I grouped them because each is the top general model its provider sells to most customers.

Workhorse and budget models

Most real apps don’t run on a flagship. They run on the next tier down, where the spread is wider.

A few things stand out:

• The $2/$10 tier is crowded. GPT-6.1 Sol and Claude Sonnet 5.5 have the same list price. Over 272K input tokens, Sol goes to $4 and $15.
• Grok Build 0.1 is $1 in and $2 out. Among the models priced at about $1 input, that is the lowest output price here.
• Gemini 3.8 Flash is $0.75 and $3.75 through December 31, 2026. It doubles to $1.50 and $7.50 on January 1, 2027.
• Kimi K2 models are $0.95 and $4, but Moonshot closes K2 to new users from October 31, 2026 and retires it by the end of December. Plan on K3.
• Haiku 5.5 is $0.10 and $0.50 only for prompts up to 100K tokens. Above that it is $0.50 and $2.50.

I covered the Kimi numbers in more detail in Kimi K3.1 Pro API pricing vs Claude and DeepSeek. For DeepSeek Flash inside a coding tool, see DeepSeek V4.1 Flash on OpenCode.

What a month actually costs

Per-token prices are hard to compare in your head. So I used one fixed workload: 100 million input tokens and 20 million output tokens a month. That is a round number you can scale up or down. No cache, no batch, short prompts.

LLM API pricing comparison for one workload

The spread is about 100 times. GPT-6 Astra and Claude Fable 5.1 cost $2,000. Claude Opus 5.5 costs $800. The $2/$10 group lands at $400. Grok 4.7 is $320. DeepSeek V4 Pro is $211 at peak. GPT-6.1 Terra is $200. Gemini 3.8 Flash is $150. At the bottom, Flash-Lite and DeepSeek Flash are around $55, and GPT-6 Luna and Haiku 5.5 are $20.

Two warnings. First, token counts are not equal across providers. Anthropic says its tokenizer for Claude 4.7 and later produces about 30% more tokens for the same text. So the same job on Sonnet 5.5 will likely cost more than the $400 shown. Second, the Argon and 3.8 Flash bars are promotional prices. After their end dates they double to $800 and $300.

Discounts that change the bill

The list price is where the bill starts. These levers move it the most.

• Batch. OpenAI (Batch and Flex), Anthropic and Google all take 50% off for work that can wait.
• Prompt caching. Cached input is 10% of the input price at Anthropic and Kimi, and 5–10% at OpenAI. xAI charges 25%. DeepSeek charges about 2%: $0.006 against $0.30 on Flash.
• Off-peak. DeepSeek’s peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday. Outside them, everything is half price.
• Surcharges. OpenAI’s Fast mode costs 2x. Anthropic’s US-only inference costs 1.1x, and OpenAI’s data residency adds 10%.
• Long prompts. OpenAI raises the price over 272K input tokens. Gemini 3.1 Pro and xAI do it over 200K. At xAI, crossing the line bills the whole request at the higher rate.

I did not check batch pricing for xAI, Kimi or DeepSeek for this article. If you need batch there, read their docs first.

Tools cost extra

Web search is billed on its own. OpenAI and Anthropic both charge $10 per 1,000 searches. Google gives 5,000 grounded searches free each month, then charges $14 per 1,000. DeepSeek charges $0.75 per 1,000 for Search and for Fetch. It is free inside Claude Code, Codex and its Harness until billing starts on November 1, 2026 (Beijing time).

Which one to pick

• You want the strongest model and cost is secondary: GPT-6 Astra or Claude Fable 5.1. Budget $10/$50.
• You want one good default: anything in the $2/$10 tier. Try two of them on your own prompts. Token counts and answer length differ, and that matters more than the list price.
• You run high volume on simple tasks: GPT-6 Luna, Haiku 5.5, Flash-Lite or DeepSeek Flash.
• Your jobs can wait: use batch at OpenAI, Anthropic or Google, or DeepSeek’s off-peak hours.

If you’re weighing an API against a flat subscription, I compared the two for OpenAI in ChatGPT subscription vs API. For a free start on Google, see Gemini API free tier limits.

What I couldn’t confirm

• Mistral is missing. Its page I checked only showed subscription plans, not API token prices.
• Promo end dates. OpenAI says GPT-5.6 Sol’s promotional price runs “at least through November 21, 2026”. It doesn’t say what happens after.
• Rate limits vary by account tier at every provider. This article compares price only.

Prices change often. Treat these numbers as a snapshot of October 27, and check the linked pages before you commit to a budget.