CodeKitHub
Cheapest LLM API in 2026: full pricing comparison, 19 models
Published Jul 25, 2026
If you’re picking an LLM API purely on cost, the spread is bigger than most people expect. Comparing all 19 models this site’s LLM API Pricing Calculator tracks — OpenAI, Anthropic, Google, DeepSeek, Qwen, Moonshot/Kimi, Zhipu/GLM and xAI — the cheapest and most expensive model differ by more than 60x on a like-for-like basis. That gap matters a lot more than which vendor’s logo you’re used to.
The ranking
To compare fairly, every model below is scored on a blended rate: 750,000 input tokens + 250,000 output tokens per 1M total tokens, roughly matching a typical chat-style request where the reply is shorter than the prompt. Prices are per 1M tokens, verified against each vendor’s official pricing page.
| Rank | Model | Provider | Input | Output | Blended (3:1) |
|---|---|---|---|---|---|
| 1 | Gemini 2.5 Flash-Lite | $0.10 | $0.40 | $0.175 | |
| 1 | DeepSeek V4 Flash | DeepSeek | $0.14 | $0.28 | $0.175 |
| 3 | Qwen3.6 Flash | Qwen | $0.19 | $1.13 | $0.42 |
| 4 | GPT-5.4 Nano | OpenAI | $0.20 | $1.25 | $0.46 |
| 5 | DeepSeek V4 Pro | DeepSeek | $0.44 | $0.87 | $0.54 |
| 6 | Grok Code Fast 1 | xAI | $1.00 | $2.00 | $1.25 |
| 7 | GPT-5.4 Mini | OpenAI | $0.75 | $4.50 | $1.69 |
| 8 | Claude Haiku 4.5 | Anthropic | $1.00 | $5.00 | $2.00 |
| 9 | GLM-5.2 | Zhipu | $1.40 | $4.40 | $2.15 |
| 10 | GPT-5.6 Luna | OpenAI | $1.00 | $6.00 | $2.25 |
| 11 | Grok 4.5 | xAI | $2.00 | $6.00 | $3.00 |
| 12 | Gemini 3.5 Flash | $1.50 | $9.00 | $3.38 | |
| 13 | Qwen3.7 Max | Qwen | $2.50 | $7.50 | $3.75 |
| 14 | Claude Sonnet 5 | Anthropic | $2.00 | $10.00 | $4.00 |
| 15 | Gemini 3.1 Pro Preview | $2.00 | $12.00 | $4.50 | |
| 16 | GPT-5.6 Terra | OpenAI | $2.50 | $15.00 | $5.63 |
| 17 | Kimi K3 | Moonshot | $3.00 | $15.00 | $6.00 |
| 18 | Claude Opus 5 | Anthropic | $5.00 | $25.00 | $10.00 |
| 19 | GPT-5.6 Sol | OpenAI | $5.00 | $30.00 | $11.25 |
A few things stand out. Every provider — not just one — has both a budget tier and a flagship tier several multiples apart, so “which vendor is cheapest” is the wrong question; “which specific model” is the right one. Second, the two cheapest models (Gemini 2.5 Flash-Lite and DeepSeek V4 Flash) land at exactly the same blended rate despite coming from unrelated companies — at the very low end, budget-tier pricing has converged rather than one vendor undercutting the field.
Why the cheapest model isn’t automatically the right choice
Price alone doesn’t tell you whether a model is good enough for your task. Budget-tier models are usually fine for classification, extraction, short replies, and other narrow, well-defined jobs — the kind of work where a flagship model’s extra reasoning capability goes unused. They’re a worse fit for open-ended reasoning, long multi-step agent workflows, or tasks where a wrong answer is expensive to catch later; a cheaper model that gets the task wrong and needs a retry (or a human fix) can end up costing more than a pricier model that gets it right the first time.
A reasonable default: prototype with whichever model you’re most familiar with, then once the task is well-defined, test whether a budget-tier model from the same table produces acceptably similar output — if it does, the savings compound at scale; if it doesn’t, the price difference wasn’t the real cost driver anyway.
Getting your actual number
This table uses a fixed 3:1 blend to make ranking possible, but your real input/output token split almost certainly isn’t 3:1 — a summarization task might be 10:1, a code-generation task closer to 1:2. The blended rank above is a reasonable starting point for picking candidates to test, not a substitute for computing your specific cost.
For that, use the LLM API Pricing Calculator directly: enter your actual input and output token counts and it computes the exact cost for any of these 19 models, plus a monthly estimate if you know roughly how many requests you’ll make. If you don’t know your token counts yet, the Token Counter tool gives you the real number from an actual prompt, rather than a rough guess. Both run entirely in your browser — no API key, no account, nothing uploaded.