RANKING

The cheapest LLM APIs, ranked

All 25 tracked models, sorted by input price — lowest cost per million tokens first.

This ranks every model we track by its raw input price, regardless of capability tier, so a budget model and a flagship model compete on the same list. That makes it a fast way to find a floor price, but price and capability move together for a reason — the cheapest option here is rarely the most capable one. Need a specific capability level instead of just the lowest price? Our use-case guides filter by use case. And some providers also offer a separate free tier before you pay anything at all.

ℹ️Ranked by input price — the cost per million tokens sent to the model, before output tokens are billed separately.
25 models
# Model Provider Input /1M Output /1M Context
1 Mistral $0.15 $0.60 Not published
2 Meta (via Together AI) $0.18 $0.59 1M tokens
3 OpenAI $0.20 $1.20 ~1.05M tokens
4 Google $0.25 $1.50 Not published
5 Meta (via Together AI) $0.27 $0.85 ~1.05M tokens
6 DeepSeek $0.30 $1.20 1M tokens
7 Amazon $0.30 $2.50 1M tokens
8 Alibaba $0.40 $1.60 1M tokens
9 Mistral $0.50 $1.50 Not published
10 Google $0.75 $3.75 Not published
11 Anthropic $1.00 $5.00 200K tokens
12 xAI $1.00 $2.00 256K tokens
13 xAI $1.25 $2.50 1M tokens
14 Amazon $1.25 $10.00 Not published (preview)
15 Mistral $1.50 $7.50 Not published
16 Anthropic $2.00 $10.00 1M tokens
17 OpenAI $2.00 $12.00 ~1.05M tokens
18 Google $2.00 $12.00 ≤200K tokens tier
19 xAI $2.00 $6.00 500K tokens
20 Alibaba $2.00 $6.00 1M tokens
21 Cohere $2.50 $10.00 256K tokens
22 OpenAI $4.00 $20.00 ~1.05M tokens
23 Anthropic $5.00 $25.00 1M tokens
24 Anthropic $10.00 $50.00 1M tokens
25 OpenAI $10.00 $50.00 ~1.05M tokens

Ranked on input price alone — output price, context window, and real-world quality all vary independently of this ranking. Check the model's own page before picking on price alone.

Frequently asked questions

Ranked by input price per million tokens — what you pay to send a request, before the model generates a response. Output tokens are usually billed at a different (often higher) rate, so check the Output column too before assuming the cheapest input price is the cheapest model overall for your workload.

No. A lower price on this list almost always means a smaller or less capable model — fine for simple, high-volume tasks, and a real problem for anything that needs strong reasoning. Match the model to the job first, then use price to pick between the remaining candidates.

No — every price here is the standard paid per-token rate from the provider's own pricing page, the same number shown everywhere else on PerTokens. A few providers also offer free, rate-limited, or trial access separately; that's covered on its own page, not folded into this ranking.

Every time a tracked price changes. Model prices are re-verified against each provider's official pricing page roughly every 48 hours, and this ranking is regenerated from the same dataset — see the changelog for a dated history of what moved.