Cheapest LLM API 2026: Verified Prices per 1M Tokens
Table of contents:
Quick answer: As of November 19, 2026, the cheapest LLM API by list price is Alibaba's Qwen3.7-Flash, at $0.03 per million input tokens and $0.13 per million output tokens for prompts up to 32K tokens.
Cheapest from a major US lab: OpenAI's GPT-6 Luna and Anthropic's Claude Haiku 5.5 tie at $0.10 input / $0.50 output per million tokens.
Cheapest open-weight model: Groq's GPT-OSS 20B at $0.075 / $0.30.
Cheapest for batch jobs: GPT-6 Luna or Claude Haiku 5.5 through their Batch APIs, at $0.05 / $0.25.
Watch for: long-prompt surcharges, DeepSeek's peak-hour prices, and promo prices that end soon. These can change the ranking for your workload.
Looking for the cheapest LLM API? The answer depends on what you mean by cheap.
The lowest price per token isn't always the lowest bill.
Long prompts, peak hours, reasoning tokens and expiring promos can flip the ranking.
We checked every price below against the provider's own pricing page on November 19, 2026.
Several popular comparison pages disagree on basic numbers, so we link each source.
You'll also see what real workloads cost per month, and how to cut your bill without switching providers.
If you'd rather not juggle five accounts, ScriptRun's OpenAI-compatible API gives you one key for OpenAI, Claude and Gemini models.
How LLM API pricing works
LLM API pricing is per token.
A token is a piece of a word, and 1 million tokens is roughly 750,000 English words.
You pay separately for:
-
Input tokens: everything you send, including the system prompt, chat history and documents.
-
Output tokens: everything the model writes. Output usually costs 4–6 times more than input.
-
Cached input: repeated prompt text the provider has stored. This is often 90% cheaper than fresh input.
-
Reasoning tokens: on "thinking" models, the hidden reasoning is billed as output.
Most providers also offer a Batch API at 50% off for jobs that can wait up to 24 hours.
Cheapest LLM API pricing comparison 2026
This LLM API pricing comparison shows each provider's cheapest current model at standard rates.
These are the LLM API pricing per million tokens 2026 rates, checked on November 19, 2026.
| Provider | Cheapest model | Input / 1M | Output / 1M | Cached input / 1M | Notes |
|---|---|---|---|---|---|
| Alibaba Cloud | Qwen3.7-Flash | $0.03 | $0.13 | discounted | Prompts up to 32K tokens; International deployment |
| Groq | GPT-OSS 20B (open weights) | $0.075 | $0.30 | $0.037 | 131K context |
| Mistral | Ministral 3 3B | $0.10 | $0.10 | $0.01 | Small edge model |
| OpenAI | GPT-6 Luna | $0.10 | $0.50 | $0.01 | Over 272K input: 2x input, 1.5x output |
| Anthropic | Claude Haiku 5.5 | $0.10 | $0.50 | $0.01 | Over 100K input: $0.50 / $2.50 |
| DeepSeek | V4.1 Flash (off-peak) | $0.15 | $0.60 | $0.003 | Peak hours: $0.30 / $1.20 |
| Alibaba Cloud | Qwen3.8-Flash | $0.15 | $0.47 | $0.016 | 1M context at one price |
| Mistral | Mistral Small 4 | $0.15 | $0.60 | $0.015 | 256K context |
| Gemini 3.1 Flash-Lite | $0.25 | $1.50 | $0.025 | Text, image and video input |
How we checked the cheapest LLM API prices
Price lists online often disagree, mix up model generations (GPT-5.6 Luna vs GPT-6 Luna, Claude Haiku 4.5 vs 5.5), or keep prices that have since changed.
So for every row we:
-
took the price from the provider's official pricing page or model page, not from an aggregator;
-
used standard (on-demand) rates in USD, and noted batch, peak and long-prompt rates separately;
-
flagged promotional prices with their end dates;
-
recorded the date we checked: November 19, 2026.
Prices change often. Before you commit, open the linked source for any model you plan to use.
The cheapest LLM API provider for each job
The cheapest LLM API provider overall isn't always the cheapest for your task.
Here's the cheapest AI model API for each common job.
Picking the cheapest LLM model API for the task usually saves more than switching to the cheapest AI API provider overall.
-
Lowest price overall: Qwen3.7-Flash. It's the cheapest API LLM option for short prompts, at $0.03 / $0.13. Above 32K tokens its price rises, so check your prompt sizes.
-
Cheapest from a major lab: GPT-6 Luna and Claude Haiku 5.5, both at $0.10 / $0.50. OpenAI calls Luna its "most efficient model for focused, high-volume tasks," and Anthropic positions Haiku 5.5 for classification, extraction and routing.
-
Cheapest multimodal LLM API: GPT-6 Luna accepts text and images at $0.10 / $0.50. For audio or video input, Gemini 3.1 Flash-Lite starts at $0.25 / $1.50.
-
Cheapest LLM API for AI agents 2026: use a cheap model for the sub-steps and a stronger one only to plan. In Anthropic's own test, Claude Opus 5.5 using Haiku 5.5 subagents cost $47, against $139 with Sonnet 5.5 subagents.
-
Cheapest for long documents: Qwen3.8-Flash keeps one price up to 1M tokens. GPT-6 Luna stays at standard rates up to 272K tokens.
Cheapest LLM inference APIs for open-weight models
If you want open-weight models, which you could also self-host later, these are the cheapest LLM inference APIs we verified:
-
Groq: GPT-OSS 20B at $0.075 / $0.30, and GPT-OSS 120B at $0.15 / $0.60, with very fast output. Its Batch API is 50% off.
-
Mistral: Ministral 3 3B at $0.10 / $0.10, Ministral 3 8B at $0.15 / $0.15, and Mistral Small 4 at $0.15 / $0.60.
-
DeepSeek: V4.1 Flash (open weights) from $0.15 / $0.60 off-peak.
Hosting platforms such as Together AI and DeepInfra also serve open-weight models.
Their prices change often, so compare them for the exact model you need.
What the cheapest LLM API really costs per month
Here's the formula behind any LLM API pricing calculator:
Monthly cost = (input tokens ÷ 1M × input price) + (output tokens ÷ 1M × output price)
We ran three common workloads through it at standard list prices.
#1. Support chatbot: 100,000 conversations a month
Each conversation is 2,000 input tokens and 300 output tokens, so 200M input and 30M output a month.
| Model | Monthly cost |
|---|---|
| Qwen3.7-Flash | $9.90 |
| Groq GPT-OSS 20B | $24.00 |
| GPT-6 Luna / Claude Haiku 5.5 | $35.00 |
| DeepSeek V4.1 Flash (off-peak / peak) | $48.00 / $96.00 |
| Gemini 3.1 Flash-Lite | $95.00 |
#2. The cheapest LLM API for batch processing millions of documents
Summarizing 1 million documents works out to 1,500 input and 200 output tokens each, so 1.5B input and 200M output.
Batch APIs halve the price:
| Model | Cost for 1M documents |
|---|---|
| Qwen3.7-Flash (batch discount) | $35.50 |
| Groq GPT-OSS 20B (Batch API) | $86.25 |
| GPT-6 Luna / Claude Haiku 5.5 (Batch API) | $125.00 |
| Gemini 3.1 Flash-Lite (Batch) | $337.50 |
| DeepSeek V4.1 Flash (off-peak, no batch) | $345.00 |
#3. Long-document Q&A: 10,000 requests with 120K-token prompts
That's 1.2B input and 5M output.
Here the long-prompt rules change the ranking:
| Model | Monthly cost |
|---|---|
| GPT-6 Luna (under 272K, standard rate) | $122.50 |
| Qwen3.8-Flash | $182.35 |
| DeepSeek V4.1 Flash (off-peak) | $183.00 |
| Claude Haiku 5.5 (over-100K rate) | $612.50 |
At the same list price as GPT-6 Luna, Haiku 5.5 costs five times more here, because every prompt is over 100K tokens.
Prompt caching would cut all of these sharply when the same documents repeat.
Hidden costs that change the cheapest LLM API ranking
- Long-prompt surcharges:\
- Claude Haiku 5.5 jumps to $0.50 / $2.50 above 100K tokens.\
- OpenAI charges 2x input and 1.5x output for the whole request above 272K tokens.\
- Gemini 3.1 Pro charges more above 200K tokens.
- Peak hours: DeepSeek charges double from 01–04 and 06–10 UTC on weekdays. Off-peak is half price.
- Promos that end:
- Gemini 4 Argon's prices double on December 1, 2026.\- Gemini 3.8 Flash's prices double on January 1, 2027.
Several comparison sites list these promo prices without the end dates.
-
Reasoning tokens: thinking models bill hidden reasoning as output. Google's pricing table says output "incl. thinking."
-
Cache writes: on OpenAI, writing to the cache costs 1.25x the normal input price. Reads are 90% cheaper.
-
Data residency: Anthropic adds a 1.1x multiplier for US-only inference.
-
Tool calls: some providers charge separately for built-in search. DeepSeek charges $0.75 per 1,000 Search or Fetch calls.
-
Rate limits: cheap models often have tight limits, and retries cost money. See our guide to OpenAI API rate limits.
How to cut costs on the cheapest LLM API (or any LLM API)
-
Cache repeated context. Put the fixed part of your prompt first so it can be cached. Anthropic found Claude Opus 5.5 costs 3.3 times more on a coding benchmark without caching.
-
Batch anything that can wait. OpenAI, Anthropic, Google, Groq and Alibaba all offer 50% off for batch jobs.
-
Route by difficulty. Send easy requests to a cheap model and hard ones to a stronger one. An LLM router does this automatically, with fallback when a model fails.
-
Cap output length. Output costs 4–6 times more than input, so set max tokens and ask for short answers. For more token-saving tactics, see our LLM API cost optimization guide.
-
Schedule around peak pricing. Run DeepSeek jobs off-peak to pay half.
-
Pick the right region. Qwen3.8-Flash costs $0.113 / $0.382 in Frankfurt against $0.15 / $0.47 in Singapore.
-
Keep switching cheap. Use an OpenAI-compatible API so moving to a cheaper provider means changing a base URL, not rewriting code.
Free options before you pay for the cheapest LLM API
If you're still prototyping, a free tier may cover you for a while.
Gemini, Groq, OpenRouter and Mistral all have one.
We compare them, with limits and data terms, in our guide to the free LLM API options.
This article covers paid prices only.
One LLM API key for every model
The cheapest LLM APIs change every few months, and each one means another account, another key and another bill.
If what you're after is the cheapest LLM API key setup to manage, ScriptRun gives you one key for OpenAI, Claude and Gemini models through an OpenAI-compatible endpoint.
|
Which cheapest LLM API should you choose
-
Short prompts and high volume: Qwen3.7-Flash is the cheapest LLM API on list price. Test its quality on your own data first.
-
You want a major US lab: GPT-6 Luna or Claude Haiku 5.5. Choose Luna if your prompts run between 100K and 272K tokens.
-
Open weights and speed: Groq's GPT-OSS 20B or 120B. If response time matters as much as price, see how to reduce API latency.
-
Batch jobs: any provider's Batch API at 50% off, or DeepSeek off-peak.
-
Several of the above: use one OpenAI-compatible endpoint and route each request to the cheapest LLM API that handles it well.
FAQs about Cheapest LLM API
What is the cheapest LLM API?
By list price on November 19, 2026, it's Alibaba's Qwen3.7-Flash, at $0.03 per million input tokens and $0.13 per million output tokens for prompts up to 32K tokens. From major US labs, GPT-6 Luna and Claude Haiku 5.5 are the cheapest at $0.10 / $0.50.
Which LLM API is cheapest for long prompts?
For prompts over 100K tokens, GPT-6 Luna (standard rates up to 272K) and Qwen3.8-Flash (one price up to 1M) are usually cheapest. Claude Haiku 5.5 charges $0.50 / $2.50 above 100K tokens.
Which is the cheapest AI API for high-volume work?
For jobs that can wait, use a Batch API. GPT-6 Luna and Claude Haiku 5.5 drop to $0.05 / $0.25 per million tokens. For short prompts, Qwen3.7-Flash with its batch discount is cheaper still.
What is the cheapest AI API to start with?
Start on a free tier while you prototype, then move to a low-cost model like GPT-6 Luna, Claude Haiku 5.5 or Qwen3.8-Flash. Using an OpenAI-compatible API keeps switching cheap later.
Which AI API is the cheapest for open-weight models?
Among the providers we checked, Groq's GPT-OSS 20B at $0.075 / $0.30 is the cheapest, followed by Mistral's Ministral 3 3B at $0.10 / $0.10.
Related posts
AI Agent Marketplace: 10 Best Places to Find, Run and Sell AI Agents in 2026
What an AI agent marketplace is, 10 top platforms compared on cost to try and who can sell, a safety checklist, and how to list your own agent in 2026.
Free LLM API: Which Providers Are Still Free in 2026 (Limits, Keys and Catches)
Every free LLM API worth using in 2026, checked against official docs: free tiers vs credits, limits, data terms, how to get a key, and fallback code.
OpenAI Compatible API: What It Is, Who Supports It and How to Switch Safely
An OpenAI compatible API guide: base URLs for 15 providers, what breaks when you switch, a copy-paste smoke test and a safe migration checklist.
OpenRouter Alternatives: 7 LLM Gateways Compared on Routing, Pricing and Lock-In
OpenRouter alternatives compared: fees, routing, BYOK and self-hosting across 7 LLM gateways, with a migration checklist and verified 2026 prices.








