Blended LLM API cost by input:output mix
Short answer: At a 3:1 input:output mix — (3 × input + output) ÷ 4 — the cheapest verified model is Ministral 3 3B at $0.100 per 1M blended tokens; the most expensive are GPT-6 Cyber at $40.00, a 400× gap. Rank 1 and the last rank stay the same across all four mixes; 27 of 55 models change rank between 10:1 and 1:3, because each model's output-to-input price ratio sets how much its blended price rises as the mix shifts to output.
The formula
A mix is a ratio of token counts, not prices. 3:1 means three input tokens for every output token. The blended price is the token-weighted average of the two rates:
blended = (w_in × input + w_out × output) ÷ (w_in + w_out)
For 3:1 that is (3 × input + output) ÷ 4; for 10:1 it is (10 × input + output) ÷ 11. The table uses the same verified input and output prices as the price list, with no cache or batch discount; those are separate levers, covered in the caching guide and batch guide.
| Mix | Input share | Cheapest | $/1M | Most expensive | $/1M | Gap |
|---|---|---|---|---|---|---|
| 10:1 | 90.9% | Ministral 3 3B | $0.100 | GPT-6 Cyber | $27.27 | 273× |
| 3:1 | 75.0% | Ministral 3 3B | $0.100 | GPT-6 Cyber | $40.00 | 400× |
| 1:1 | 50.0% | Ministral 3 3B | $0.100 | GPT-6 Cyber | $60.00 | 600× |
| 1:3 | 25.0% | Ministral 3 3B | $0.100 | GPT-6 Cyber | $80.00 | 800× |
Ties share a cell: GPT-6 Cyber both charge $10 input and $50 output, so their blended price is identical at every mix.
Rank by mix
Every model at every mix
Sorted by the 3:1 blended price. “Sensitivity” is the 1:3 price divided by the 10:1 price: how much the mix assumption changes that model's blended rate. Rank is computed across all 55 verified models; equal prices share a rank.
| Rank | Model | Input /1M | Output /1M | 10:1 | 3:1 | 1:1 | 1:3 | 1:3 ÷ 10:1 |
|---|---|---|---|---|---|---|---|---|
| 1 | Ministral 3 3B | $0.1 | $0.1 | $0.100 | $0.100 | $0.100 | $0.100 | 1.00× |
| 2 | Ministral 3 8B | $0.15 | $0.15 | $0.150 | $0.150 | $0.150 | $0.150 | 1.00× |
| 3 | Gemini 2.5 Flash-Lite | $0.1 | $0.4 | $0.127 | $0.175 | $0.250 | $0.325 | 2.55× |
| 4 | GPT-6 Luna | $0.1 | $0.5 | $0.136 | $0.200 | $0.300 | $0.400 | 2.93× |
| 4 | Claude Haiku 5.5 | $0.1 | $0.5 | $0.136 | $0.200 | $0.300 | $0.400 | 2.93× |
| 4 | Ministral 3 14B | $0.2 | $0.2 | $0.200 | $0.200 | $0.200 | $0.200 | 1.00× |
| 7 | GPT-4o mini | $0.15 | $0.6 | $0.191 | $0.262 | $0.375 | $0.487 | 2.55× |
| 7 | Mistral Small 4 | $0.15 | $0.6 | $0.191 | $0.262 | $0.375 | $0.487 | 2.55× |
| 9 | Codestral 25.08 | $0.3 | $0.9 | $0.355 | $0.450 | $0.600 | $0.750 | 2.12× |
| 9 | GPT-5.6 Luna | $0.2 | $1.2 | $0.291 | $0.450 | $0.700 | $0.950 | 3.27× |
| 11 | GPT-5.4 nano | $0.2 | $1.25 | $0.295 | $0.463 | $0.725 | $0.988 | 3.34× |
| 12 | DeepSeek Flash (V4.1) | $0.3 | $1.2 | $0.382 | $0.525 | $0.750 | $0.975 | 2.55× |
| 13 | Gemini 3.1 Flash-Lite | $0.25 | $1.5 | $0.364 | $0.562 | $0.875 | $1.19 | 3.27× |
| 14 | GPT-5 mini | $0.25 | $2 | $0.409 | $0.688 | $1.12 | $1.56 | 3.82× |
| 15 | Gemini 3.5 Flash-Lite | $0.3 | $2.5 | $0.500 | $0.850 | $1.40 | $1.95 | 3.90× |
| 15 | Gemini 2.5 Flash | $0.3 | $2.5 | $0.500 | $0.850 | $1.40 | $1.95 | 3.90× |
| 17 | Mistral Large 4 | $0.7 | $2.1 | $0.827 | $1.05 | $1.40 | $1.75 | 2.12× |
| 18 | Grok Build 0.1 | $1 | $2 | $1.09 | $1.25 | $1.50 | $1.75 | 1.60× |
| 19 | Gemini 3.8 Flash | $0.75 | $3.75 | $1.02 | $1.50 | $2.25 | $3.00 | 2.93× |
| 19 | Gemini 3.6 Flash | $0.75 | $3.75 | $1.02 | $1.50 | $2.25 | $3.00 | 2.93× |
| 21 | Grok 4.3 | $1.25 | $2.5 | $1.36 | $1.56 | $1.88 | $2.19 | 1.60× |
| 22 | GPT-5.4 mini | $0.75 | $4.5 | $1.09 | $1.69 | $2.62 | $3.56 | 3.27× |
| 23 | o4-mini | $1.1 | $4.4 | $1.40 | $1.93 | $2.75 | $3.58 | 2.55× |
| 24 | DeepSeek V4 Pro | $1.32 | $3.96 | $1.56 | $1.98 | $2.64 | $3.30 | 2.12× |
| 25 | GPT-6.1 Terra | $1 | $5 | $1.36 | $2.00 | $3.00 | $4.00 | 2.93× |
| 25 | Claude Haiku 4.5 | $1 | $5 | $1.36 | $2.00 | $3.00 | $4.00 | 2.93× |
| 27 | Grok 4.7 | $2 | $6 | $2.36 | $3.00 | $4.00 | $5.00 | 2.12× |
| 28 | Gemini 2.5 Pro | $1.25 | $10 | $2.05 | $3.44 | $5.62 | $7.81 | 3.82× |
| 29 | GPT-4.1 | $2 | $8 | $2.55 | $3.50 | $5.00 | $6.50 | 2.55× |
| 30 | GPT-6.1 Sol | $2 | $10 | $2.73 | $4.00 | $6.00 | $8.00 | 2.93× |
| 30 | GPT-6 Sol | $2 | $10 | $2.73 | $4.00 | $6.00 | $8.00 | 2.93× |
| 30 | Claude Sonnet 5.5 | $2 | $10 | $2.73 | $4.00 | $6.00 | $8.00 | 2.93× |
| 30 | Claude Sonnet 5 | $2 | $10 | $2.73 | $4.00 | $6.00 | $8.00 | 2.93× |
| 30 | Gemini 4 Carbon | $2 | $10 | $2.73 | $4.00 | $6.00 | $8.00 | 2.93× |
| 30 | Gemini 4 Argon | $2 | $10 | $2.73 | $4.00 | $6.00 | $8.00 | 2.93× |
| 36 | GPT-5.6 Terra | $2 | $12 | $2.91 | $4.50 | $7.00 | $9.50 | 3.27× |
| 36 | Gemini 3.1 Pro Preview | $2 | $12 | $2.91 | $4.50 | $7.00 | $9.50 | 3.27× |
| 38 | GPT-5.4 | $2.5 | $15 | $3.64 | $5.62 | $8.75 | $11.88 | 3.27× |
| 39 | Claude Sonnet 4.6 | $3 | $15 | $4.09 | $6.00 | $9.00 | $12.00 | 2.93× |
| 40 | GPT-5.6 Sol | $4 | $20 | $5.45 | $8.00 | $12.00 | $16.00 | 2.93× |
| 40 | Claude Opus 5.5 | $4 | $20 | $5.45 | $8.00 | $12.00 | $16.00 | 2.93× |
| 42 | GPT-5.5 Rosalind Research | $5 | $25 | $6.82 | $10.00 | $15.00 | $20.00 | 2.93× |
| 42 | GPT-5.5 Rosalind Discovery | $5 | $25 | $6.82 | $10.00 | $15.00 | $20.00 | 2.93× |
| 42 | Claude Opus 5 | $5 | $25 | $6.82 | $10.00 | $15.00 | $20.00 | 2.93× |
| 45 | GPT-5.5 | $5 | $30 | $7.27 | $11.25 | $17.50 | $23.75 | 3.27× |
| 46 | Claude Fable 5.2 | $8 | $40 | $10.91 | $16.00 | $24.00 | $32.00 | 2.93× |
| 46 | Claude Mythos 5.2 | $8 | $40 | $10.91 | $16.00 | $24.00 | $32.00 | 2.93× |
| 48 | GPT-6 Astra | $10 | $50 | $13.64 | $20.00 | $30.00 | $40.00 | 2.93× |
| 48 | GPT-6 Rosalind | $10 | $50 | $13.64 | $20.00 | $30.00 | $40.00 | 2.93× |
| 48 | Claude Fable 5.1 | $10 | $50 | $13.64 | $20.00 | $30.00 | $40.00 | 2.93× |
| 48 | Claude Mythos 5.1 | $10 | $50 | $13.64 | $20.00 | $30.00 | $40.00 | 2.93× |
| 48 | Claude Fable 5 | $10 | $50 | $13.64 | $20.00 | $30.00 | $40.00 | 2.93× |
| 48 | Claude Mythos 5 | $10 | $50 | $13.64 | $20.00 | $30.00 | $40.00 | 2.93× |
| 54 | GPT-5.6 Cyber | $12.5 | $75 | $18.18 | $28.12 | $43.75 | $59.38 | 3.27× |
| 55 | GPT-6 Cyber | $20 | $100 | $27.27 | $40.00 | $60.00 | $80.00 | 2.93× |
How the mix changes the ranking
27 of 55 models change rank between 10:1 and 1:3. The extremes do not move: Ministral 3 3B has the lowest output price outright and the joint-lowest input price, so it is cheapest at every mix; the two $10/$50 models are highest on both rates. What moves is the middle.
Dividing a model's 1:3 price by its 10:1 price cancels its input rate and leaves a factor that depends only on r = output price ÷ input price:
factor = 11 × (1 + 3r) ÷ (4 × (10 + r))
Across the verified set the factor runs from 1.00× at r = 1 (Ministral 3 14B, Ministral 3 8B, Ministral 3 3B) to 3.90× at r = 8.33 (Gemini 3.5 Flash-Lite, Gemini 2.5 Flash). A low ratio means output is not much dearer than input, so an output-heavy mix barely changes that model's blended price; a high ratio means the opposite. That is why the mid-table order is mix-dependent even though the cheapest model is not.
| Model | Output ÷ input | Rank at 10:1 | Rank at 1:3 | Move |
|---|---|---|---|---|
| Grok Build 0.1 | 2.00× | 20 | 15 | Up 5 |
| Ministral 3 14B | 1.00× | 8 | 3 | Up 5 |
| DeepSeek V4 Pro | 3.00× | 26 | 22 | Up 4 |
| Claude Haiku 4.5 | 5.00× | 22 | 25 | Down 3 |
| GPT-5.4 mini | 6.00× | 20 | 23 | Down 3 |
| GPT-6.1 Terra | 5.00× | 22 | 25 | Down 3 |
| Grok 4.3 | 2.00× | 22 | 19 | Up 3 |
| Ministral 3 8B | 1.00× | 5 | 2 | Up 3 |
| Claude Haiku 5.5 | 5.00× | 3 | 5 | Down 2 |
| Codestral 25.08 | 3.00× | 11 | 9 | Up 2 |
| DeepSeek Flash (V4.1) | 4.00× | 13 | 11 | Up 2 |
| GPT-5.4 nano | 6.25× | 10 | 12 | Down 2 |
| GPT-6 Luna | 5.00× | 3 | 5 | Down 2 |
| Gemini 2.5 Flash | 8.33× | 15 | 17 | Down 2 |
| Gemini 2.5 Flash-Lite | 4.00× | 2 | 4 | Down 2 |
| Gemini 2.5 Pro | 8.00× | 27 | 29 | Down 2 |
| Gemini 3.5 Flash-Lite | 8.33× | 15 | 17 | Down 2 |
| Gemini 3.6 Flash | 5.00× | 18 | 20 | Down 2 |
| Gemini 3.8 Flash | 5.00× | 18 | 20 | Down 2 |
| Mistral Large 4 | 3.00× | 17 | 15 | Up 2 |
Example: Grok Build 0.1 charges $1 input and $2 output (r = 2), so its blended price rises only 1.60× and it moves from rank 20 to 15. Gemini 3.5 Flash-Lite charges $0.3 and $2.5 (r ≈ 8.33); its blended price rises 3.90× and it slips from rank 15 to 17.
Use your own token counts
The four mixes cover common shapes but not every workload. A 2,000-input / 500-output support reply is 4:1, close to 3:1; a 20,000 / 1,000 RAG answer is 20:1, more input-heavy than 10:1; a 500 / 4,000 long-form answer is 1:8, more output-heavy than 1:3. Log 50–100 real requests, average both token fields the API returns, and use the calculator for the exact blended price. The same token counts feed the self-host break-even and the per-user budget.
Worked example: from a usage export to a blended price
Suppose last month's usage export shows 42 million input tokens and 9 million output tokens. The mix is 42 ÷ 9 = 4.67:1 — closer to 3:1 than 10:1 — and the input share is 42 ÷ 51 = 82.4%.
| Model | Blended $/1M at 4.67:1 | Monthly bill | Check: input + output billed separately |
|---|---|---|---|
| Gemini 3.8 Flash | (42 × $0.75 + 9 × $3.75) ÷ 51 = $1.279 | $1.279 × 51 = $65.25 | $31.50 + $33.75 = $65.25 |
| GPT-5.4 mini | (42 × $0.75 + 9 × $4.50) ÷ 51 = $1.412 | $1.412 × 51 = $72.00 | $31.50 + $40.50 = $72.00 |
| Claude Sonnet 5.5 | (42 × $2.00 + 9 × $10.00) ÷ 51 = $3.412 | $3.412 × 51 = $174.00 | $84.00 + $90.00 = $174.00 |
| Grok 4.3 | (42 × $1.25 + 9 × $2.50) ÷ 51 = $1.471 | $1.471 × 51 = $75.00 | $52.50 + $22.50 = $75.00 |
The last two columns always agree: a blended price is just the bill divided by total tokens, so it is useful for comparing models at your own mix, not for discovering savings. Grok 4.3 looks expensive on input but its low output price pulls its blended rate down as the mix moves toward output. Prices are list rates from each provider's pricing page (see the price list for links and check dates), before caching or batch discounts.
Questions
What is a blended API price?
A blended price is a weighted average of the input and output rates: (input tokens × input price + output tokens × output price) ÷ total tokens. At a 3:1 input:output mix this is (3 × input + output) ÷ 4 per 1M tokens. It compares models before you know your exact token counts.
Which mix should I use?
Measure input and output tokens from real requests and divide them. A 2,000-input / 500-output call is a 4:1 mix; 20,000 / 1,000 is 20:1; 500 / 4,000 is 1:8. Those do not sit exactly on the four mixes here, so use the calculator with your own counts when a few percent decides the choice.
Why does the ranking change with the mix?
Every model's blended price rises as output takes a larger share, but a model with a low output-to-input price ratio rises less. Dividing the 1:3 price by the 10:1 price gives a factor from 1.60× to 3.90× across the verified models, enough to move several mid-table models by 2–4 places.
Does the blended price include caching or batch discounts?
No. The table uses standard list input and output rates, no cached input and no batch discount. Cached input lowers the input side; batch rates are about 50% of standard on OpenAI, Anthropic and Google and 20% on the xAI Grok models that support batch. Both are separate inputs in the calculator and the guides.
Where do the prices come from?
The official provider pricing pages linked on every model page. Each row shows the date that model was checked, and the footer shows the most recent check. If a price cannot be verified it is marked unverified and excluded from totals. The verification rules are on the method page.
Next step → Compare the same API price with a self-hosted GPU
Price data last checked 2026-11-18 against official sources · how we verify
Price-change alerts
One short email when a provider changes a price or a promo ends. No spam, unsubscribe any time.