AI API Cost Comparison
Cheapest LLM APIs for Side Projects: Real Monthly Costs (October 2026)
A same-workload ranking of eight active text LLM APIs, with monthly costs at three request volumes, official sources, and verification dates.
Ministral 3 3B is the cheapest listed text LLM API for this workload: input $0.10 per 1M tokens, output $0.10 per 1M tokens, about $1.10 per month for 10,000 requests of 800 input and 300 output tokens. Ministral 3 8B ($1.65) and Ministral 3 14B ($2.20) follow. Prices verified Oct 9, 2026.
| Model | Input $/1M | Cached input $/1M | Output $/1M | Source | Last verified |
|---|---|---|---|---|---|
| Ministral 3 3B | $0.10 | $0.01 | $0.10 | Official source | Oct 9, 2026 |
| Ministral 3 8B | $0.15 | $0.015 | $0.15 | Official source | Oct 9, 2026 |
| Ministral 3 14B | $0.20 | $0.02 | $0.20 | Official source | Oct 9, 2026 |
Prices from official provider pages, verified on the date shown.
10 min read - Published 2026-10-21 - Updated 2026-10-21
The cheapest APIs for one disclosed side-project workload
The ranking starts with the site's curated, default-safe API calculator list in the production pricing mode. It excludes preview, deprecated, retired, and non-text models, requires an official source and verification date, then sorts active candidates by estimated monthly cost at 10,000 requests.
Every row uses the same assumption: 800 input tokens plus 300 output tokens per request, Standard token pricing, no cached input, retries, tools, batch discount, long-context tier, or other billable components. Monthly cost is requests multiplied by input and output token cost.
| Model | Provider | Input $/1M | Output $/1M | 1,000 requests | 10,000 requests | 100,000 requests | Source | Last verified |
|---|---|---|---|---|---|---|---|---|
| Ministral 3 3B | Mistral AI | $0.10 | $0.10 | $0.11 | $1.10 | $11.00 | Official source | Oct 9, 2026 |
| Ministral 3 8B | Mistral AI | $0.15 | $0.15 | $0.165 | $1.65 | $16.50 | Official source | Oct 9, 2026 |
| Ministral 3 14B | Mistral AI | $0.20 | $0.20 | $0.22 | $2.20 | $22.00 | Official source | Oct 9, 2026 |
| Claude Haiku 5.5 | Anthropic | $0.10 | $0.50 | $0.23 | $2.30 | $23.00 | Official source | Oct 9, 2026 |
| GPT-6 Luna | OpenAI | $0.10 | $0.50 | $0.23 | $2.30 | $23.00 | Official source | Sep 23, 2026 |
| Mistral Small 4 | Mistral AI | $0.15 | $0.60 | $0.30 | $3.00 | $30.00 | Official source | Oct 9, 2026 |
| Codestral | Mistral AI | $0.30 | $0.90 | $0.51 | $5.10 | $51.00 | Official source | Oct 9, 2026 |
| GPT-5.6 Luna | OpenAI | $0.20 | $1.20 | $0.52 | $5.20 | $52.00 | Official source | Sep 29, 2026 |
The first row is cheapest only for this equal-token workload. A different input/output mix, cache share, model-specific tier, batch mode, retry rate, or successful-task rate can change the practical choice.
Free options first
Google offers limited free input and output for certain eligible models; check the sourced model list and project-limit guidance in the Gemini API Free tier guide.
Anthropic says new users receive a small amount of credits to test the API, but it publishes no amount in the pricing FAQ; read the Claude API starter credits guide.
OpenAI documents a usage-capped Free tier for users in allowed geographies, not unrestricted access or free money; read the free OpenAI API key guide.
Cheapest per token does not mean cheapest per successful task
This page owns the equal-workload "cheapest LLM API" decision. For quality-adjusted economics, read why the cheapest LLM can cost more per successful task and use the LLM retry and fallback cost calculator with observed retries, fallbacks, and accepted outcomes.
How to choose from the shortlist
- Run the same representative tasks through the top candidates and define an acceptance check before comparing results.
- Measure actual input, output, cached input, retries, fallbacks, latency, and any tool charges instead of assuming every request has the same shape.
- Set a provider-side spend limit or alert before moving from a prototype to public traffic.
- Recheck the official source and verification date when a provider changes a model, price, lifecycle state, or access policy.
FAQ
What is the cheapest LLM API?
For this page, it is the lowest-cost active text model in the curated calculator list at 10,000 requests with 800 input and 300 output tokens each. The ranking is recalculated from current verified prices, so the first row is the answer for that exact workload.
How much does an LLM API cost for a side project?
Multiply requests by the cost of input and output tokens per request. This guide shows 1,000, 10,000, and 100,000 monthly requests using 800 input and 300 output tokens, before caching, retries, tools, or batch discounts.
Are free LLM APIs cheaper than the ranked paid models?
They can be while your project remains eligible and inside current limits. Free access can restrict models, capacity, features, or data handling, so check the linked official page and compare the paid fallback before launch.
Why can the cheapest token price cost more per task?
A low token rate can lose its advantage when the workflow uses longer responses, more retries, fallback calls, or lower task acceptance. Compare measured cost per successful task separately from this equal-token ranking.
Does this ranking include preview or retired models?
No. It starts from the curated, default-safe API calculator list and excludes preview, deprecated, retired, and non-text candidates. Every displayed row must have an official source and a last-verified date.
Replace the comparison workload with your own
Enter your model, token counts, request volume, and active users to estimate the side project you are actually building.