Which LLM API is the cheapest per million tokens?

For most everyday text workloads the cheapest options are open-weight and budget models, DeepSeek V4 Flash ($0.14 input / $0.28 output per 1M tokens), Grok 4 Fast and Gemini 2.5 Flash-Lite all sit well under $0.50 per 1M output tokens. The exact cheapest depends on your input-to-output ratio, so enter your own volumes above to see the true ranking for your workload.

How much does the GPT API cost per month?

It depends entirely on token volume. As a worked example, a support chatbot using 2,000,000 input and 500,000 output tokens a month on GPT-5.5 (at $5/$30 per 1M) costs (2 × $5) + (0.5 × $30) = $25.00/mo. The same workload on GPT-5 nano costs about $0.30/mo. Enter your figures above for an exact monthly total.

Is Claude or Gemini cheaper for output tokens?

Gemini is usually cheaper on output. Gemini 2.5 Flash charges $2.50 per 1M output tokens versus Claude Sonnet 4.5 at $15 and Claude Opus 4.8 at $25. Output tokens dominate most bills because models generate far more than a fixed system prompt, so the output rate is where the biggest savings sit.

What is the difference between input and output token pricing?

Input tokens are everything you send, the system prompt, chat history and user message. Output tokens are what the model generates back. Output is almost always billed at a higher rate (often 4–6× the input rate) because generation is more compute-intensive. That is why a chatty assistant with long replies costs more than a classifier that returns one word.

How do I estimate my monthly OpenAI or Anthropic bill?

Multiply your monthly input tokens by the input rate and your monthly output tokens by the output rate, both per 1,000,000 tokens, then add them. This tool does exactly that for 30+ models at once. If you use prompt caching, set the cached share above and cached tokens are billed at the provider's cache-read rate instead.

How does prompt caching change the price?

Providers like OpenAI, Anthropic and Google bill repeated (cached) input tokens at a discounted cache-read rate, often around one-tenth of the standard input rate. Set your cached input share above and the calculator splits input into cached and uncached buckets. Models with no published cache rate charge the full input price for cached tokens.

Why show prices in LKR, and what rate is used?

All providers bill in US dollars, so a Sri Lankan developer's real cost depends on the USD→LKR rate their card or bank applies. Switch the currency toggle to LKR and the whole table converts at the editable rate (default Rs 305, the CBSL indicative figure). Change it to match your bank's actual conversion rate for an accurate rupee estimate.

Are these prices live, and how current are they?

No, to stay fast and reliable this tool uses a static price table verified against each provider's official pricing page on 2026-07-17, not a live feed. LLM prices change often, so treat the figures as a close estimate and confirm the current rate on the provider's page (linked in Sources) before committing to a model.

Does the ranking mean the cheapest model is the best choice?

No. This tool ranks on price only. A budget model may not match a frontier model on reasoning, coding or long-context quality. Use the price ranking to shortlist affordable options, then compare capability separately, the AI Model Comparison tool covers features and benchmarks.

AI · Developer tools

LLM API Pricing Comparison

Enter your monthly input and output token volume and see the exact per-month API cost of 26 text models, GPT, Claude, Gemini, Grok, DeepSeek, Mistral and Llama, ranked cheapest-first, with a USD→LKR toggle. No signup, no ads.

By Induwara Ashinsana· Co-Founder, Ryzera TechnologiesUpdated Jul 17, 2026

Compare LLM API prices26 models

Prices verified 2026-07-17

Monthly input tokens

1,000,000 tokens/month, Prompt tokens you send per month (whole number).

Monthly output tokens

250,000 tokens/month, Completion tokens the model returns per month.

Example workloads

Cached input share (%)

Currency

Provider rates are in USD; LKR uses the rate on the right.

USD → LKR rate

CBSL indicative default, edit to your bank's rate.

Tiers

Cheapest for your workload

Mistral Ministral 8B$0.125/mo

#	Model	In $/1M	Out $/1M	Monthly cost
1	Ministral 8BBest value Mistral · Budget	$0.10	$0.10	$0.125
2	GPT-5 nano OpenAI · Budget	$0.05	$0.40	$0.15
3	Mistral Small Mistral · Budget	$0.10	$0.30	$0.175
4	Llama 4 Scout Meta · Open-weight	$0.11	$0.34	$0.195
5	Gemini 2.5 Flash-Lite Google · Budget	$0.10	$0.40	$0.20
6	DeepSeek V4 Flash DeepSeek · Budget	$0.14	$0.28	$0.21
7	Grok 4 Fast xAI · Budget	$0.20	$0.50	$0.325
8	Llama 4 Maverick Meta · Open-weight	$0.20	$0.60	$0.35
9	Qwen3 235B Alibaba · Open-weight	$0.20	$0.60	$0.35
10	DeepSeek V4 DeepSeek · Budget	$0.28	$0.42	$0.385
11	GPT-5.4 mini OpenAI · Mid-tier	$0.25	$2.00	$0.75
12	Mistral Medium Mistral · Mid-tier	$0.40	$2.00	$0.90
13	Gemini 2.5 Flash Google · Mid-tier	$0.30	$2.50	$0.925
14	DeepSeek R2 DeepSeek · Mid-tier	$0.55	$2.19	$1.0975
15	Llama 3.3 70B Meta · Open-weight	$0.88	$0.88	$1.10
16	Gemini 3 Flash Google · Mid-tier	$0.50	$3.00	$1.25
17	Claude Haiku 4.5 Anthropic · Mid-tier	$1.00	$5.00	$2.25
18	Qwen3 Max Alibaba · Frontier	$1.20	$6.00	$2.70
19	Mistral Large Mistral · Mid-tier	$2.00	$6.00	$3.50
20	Command A Cohere · Mid-tier	$2.50	$10.00	$5.00
21	Gemini 3.1 Pro Google · Frontier	$2.00	$12.00	$5.00
22	GPT-5.4 OpenAI · Frontier	$2.50	$15.00	$6.25
23	Claude Sonnet 4.5 Anthropic · Frontier	$3.00	$15.00	$6.75
24	Grok 4.1 xAI · Frontier	$3.00	$15.00	$6.75
25	Claude Opus 4.8 Anthropic · Frontier	$5.00	$25.00	$11.25
26	GPT-5.5 OpenAI · Frontier	$5.00	$30.00	$12.50

Rates are each provider's published USD price per 1,000,000 tokens for standard synchronous text, last verified 2026-07-17. Figures exclude batch, fine-tuning, image, audio and cache-write charges. Always confirm against the provider's pricing page before committing, see sources below.

How it works

The calculator answers one question, which LLM API is cheapest for my workload, by applying each provider's published per-token rate to the exact volume you enter. Every model's price is stored as USD per 1,000,000 tokens for standard synchronous text, split into a separate input (prompt) rate and output (completion) rate, because providers bill the two differently and output is usually the larger cost.

For a model with input price pin, output price pout and cache-read price pcache, given monthly input tokens I, output tokens O and a cached input share s:

Split input into cached and uncached tokens: uncached = I × (1 − s) and cached = I × s.
Input cost = (uncached ÷ 1,000,000 × pin) + (cached ÷ 1,000,000 × pcache). Models with no published cache rate use pcache = pin (no discount).
Output cost = O ÷ 1,000,000 × pout.
Monthly cost (USD) = input cost + output cost. The table sorts ascending by this figure and flags the minimum as Best value.
If you switch to LKR, every figure is multiplied by your editable USD→LKR rate (default Rs 305, the Central Bank of Sri Lanka indicative rate).

The result is cross-checked by an independent, algebraically-equivalent blended-rate formula that folds the cached and uncached input rates into one effective per-token price, effInput = pin × (1 − s) + pcache × s , and produces the same total to the cent. Prices are a static, verified table (last checked 2026-07-17); no live network calls are made, so the tool stays fast and never breaks on a failed price fetch. Because model prices change often, always confirm the current rate on the provider's page before committing, every source is linked below.

Worked examples

Support chatbot, 2,000,000 input + 500,000 output tokens/mo, 0% cached

GPT-5.5: (2 × $5) + (0.5 × $30) = $10.00 + $15.00 = $25.00/mo
Claude Sonnet: (2 × $3) + (0.5 × $15) = $6.00 + $7.50 = $13.50/mo
Gemini 2.5 Flash: (2 × $0.30) + (0.5 × $2.50) = $0.60 + $1.25 = $1.85/mo
DeepSeek V4 Flash: (2 × $0.14) + (0.5 × $0.28) = $0.28 + $0.14 = $0.42/mo
Cheapest → dearest: DeepSeek $0.42 · Gemini $1.85 · Sonnet $13.50 · GPT-5.5 $25.00
In LKR @305: DeepSeek ≈ Rs 128.10/mo · GPT-5.5 ≈ Rs 7,625.00/mo

Coding assistant, 5,000,000 input + 3,000,000 output tokens/mo, 0% cached

Gemini 3.1 Pro: (5 × $2) + (3 × $12) = $10.00 + $36.00 = $46.00/mo
GPT-5.4: (5 × $2.50) + (3 × $15) = $12.50 + $45.00 = $57.50/mo
Claude Opus 4.8: (5 × $5) + (3 × $25) = $25.00 + $75.00 = $100.00/mo
Ranked: Gemini 3.1 Pro $46.00 · GPT-5.4 $57.50 · Claude Opus 4.8 $100.00
Cheapest verdict: Gemini 3.1 Pro ≈ Rs 14,030/mo at 305

Cached edge case, 1,000,000 input, 0 output, 100% cached (DeepSeek V4 Flash)

Cache-read rate: $0.014 per 1M (vs $0.14 standard input)
uncached = 1,000,000 × (1 − 1.00) = 0 tokens
cached = 1,000,000 × 1.00 = 1,000,000 tokens
Input cost: (0 × $0.14) + (1 × $0.014) = $0.014/mo
Output cost: 0 → Monthly cost = $0.014/mo (full cache discount applied)

Frequently asked questions

Sources & references

Every rate in the price table was cross-checked against these official pricing pages on 2026-07-17. Provider-hosted open-weight models (Llama, Qwen) are priced by the serving provider, not the model author. LLM prices change frequently, confirm the current figure on the provider's page before you commit to a model.

Related tools

AI API Cost Calculator

Estimate the monthly and per-request USD and LKR bill for any major LLM API. Pick a model, enter input/output tokens and requests per month, and compare every model cheapest-first, with optional 50% batch and cached-input discounts.

Open tool

LLM Price History Tracker

Track the dated API price of every major large-language model, GPT, Claude, Gemini, Grok and hosted open-weights, from launch to today, in one filterable table and trend chart. A cost time-machine shows what your monthly token workload would have cost in 2023, 2024, 2025 and now, and exactly how much you save today. Every figure is a cited, dated list price.

Open tool

LLM Speed Comparison

Compare the real-world response speed of hosted LLM APIs, GPT, Claude, Gemini, Llama, DeepSeek, Grok, Mistral. Pick 2–6 models, enter your output length, and rank them by estimated total response time, output tokens/sec, and time-to-first-token. Every figure cited from Artificial Analysis medians.

Open tool

Comments & feedback

Spotted a bug or want an improvement? Tell us, our team reviews every comment, and good ideas get built. Comments are public and anonymous.

Spotted a stale price, missing model, or edge case?

Email me at [email protected] , most fixes ship within 24 hours.