Benchmark · as of 30 September 2026

AI Cost Benchmark 2026: What One Task Costs in AI Fees

46 current models, 12 business scenarios, vendor list prices. The tables show what a single task costs in AI fees in euros, computed with the engine of the AI cost calculator, version 10.

46
current models
12
business scenarios
6
vendors and hosts

The range per scenario

Depending on the scenario, the most expensive current model costs 374 to 925 times as much as the cheapest. The median shows where the field stands; the factor shows how much model choice matters.

The range per scenario
Scenario	Models	cheapest model	Median	most expensive model	Factor
Answering a support ticket	46	Ministral 3 3B 0.0004 €	0.0057 €	gpt-5.4-pro 0.1612 €	403×
Reading an email and drafting the reply	46	Ministral 3 3B 0.0002 €	0.0034 €	gpt-5.4-pro 0.1004 €	502×
Writing or revising a text	46	Ministral 3 3B 0.0005 €	0.0088 €	gpt-5.4-pro 0.2959 €	592×
Sales email and proposal draft	46	Ministral 3 3B 0.0006 €	0.0112 €	gpt-5.4-pro 0.3487 €	581×
Classification and routing	46	Ministral 3 3B 0.0001 €	0.0018 €	gpt-5.4-pro 0.0444 €	444×
Extracting data from documents	46	Qwen3.8 Flash (Together) 0.0013 €	0.0195 €	gpt-5.4-pro 0.4914 €	378×
Knowledge-base question	46	Ministral 3 3B 0.0007 €	0.0107 €	gpt-5.4-pro 0.2748 €	393×
Research agent over documents	46	Qwen3.8 Flash (Together) 0.0182 €	0.2660 €	gpt-5.4-pro 6.8058 €	374×
Writing or changing code	46	Ministral 3 3B 0.0044 €	0.0680 €	gpt-5.4-pro 1.9022 €	432×
Working through a long document	38	Qwen3.8 Flash (Together) 0.0245 €	0.3949 €	gpt-5.4-pro 16.5654 €	676×
Meeting notes	46	Ministral 3 3B 0.0014 €	0.0202 €	gpt-5.4-pro 0.5231 €	374×
Translation	46	Ministral 3 3B 0.0004 €	0.0098 €	gpt-5.4-pro 0.3699 €	925×

The cheapest models per scenario

For each scenario the five cheapest models by AI fees per task, plus the cheapest model in each class. The input sizes are listed below each table and come from the calculator presets.

Answering a support ticket

Input 3,700 tokens, output 400 tokens, one call per task

Answering a support ticket
Model	Class	AI fees per task
Ministral 3 3B	Workhorse class	0.0004 €
Qwen3.8 Flash (Together)	Workhorse class	0.0004 €
Ministral 3 8B	Workhorse class	0.0005 €
gpt-6-luna	Compact class	0.0005 €
DeepSeek V4 Flash 0731 (Together)	Workhorse class	0.0006 €
Cheapest model per class
Mistral Large 3	Frontier class	0.0022 €
Ministral 3 3B	Workhorse class	0.0004 €
gpt-6-luna	Compact class	0.0005 €

Reading an email and drafting the reply

Input 2,000 tokens, output 300 tokens, one call per task

Reading an email and drafting the reply
Model	Class	AI fees per task
Ministral 3 3B	Workhorse class	0.0002 €
Qwen3.8 Flash (Together)	Workhorse class	0.0002 €
Ministral 3 8B	Workhorse class	0.0003 €
DeepSeek V4 Flash 0731 (Together)	Workhorse class	0.0003 €
gpt-6-luna	Compact class	0.0003 €
Cheapest model per class
Mistral Large 3	Frontier class	0.0013 €
Ministral 3 3B	Workhorse class	0.0002 €
gpt-6-luna	Compact class	0.0003 €

Writing or revising a text

Input 4,000 tokens, output 1,200 tokens, one call per task

Writing or revising a text
Model	Class	AI fees per task
Ministral 3 3B	Workhorse class	0.0005 €
Qwen3.8 Flash (Together)	Workhorse class	0.0006 €
Ministral 3 8B	Workhorse class	0.0007 €
DeepSeek V4 Flash 0731 (Together)	Workhorse class	0.0008 €
Ministral 3 14B	Workhorse class	0.0009 €
Cheapest model per class
Mistral Large 3	Frontier class	0.0033 €
Ministral 3 3B	Workhorse class	0.0005 €
gpt-6-luna	Compact class	0.0009 €

Sales email and proposal draft

Input 6,000 tokens, output 1,200 tokens, one call per task

Sales email and proposal draft
Model	Class	AI fees per task
Ministral 3 3B	Workhorse class	0.0006 €
Qwen3.8 Flash (Together)	Workhorse class	0.0008 €
Ministral 3 8B	Workhorse class	0.0010 €
DeepSeek V4 Flash 0731 (Together)	Workhorse class	0.0010 €
gpt-6-luna	Compact class	0.0011 €
Cheapest model per class
Mistral Large 3	Frontier class	0.0042 €
Ministral 3 3B	Workhorse class	0.0006 €
gpt-6-luna	Compact class	0.0011 €

Classification and routing

Input 1,500 tokens, output 30 tokens, one call per task

Classification and routing
Model	Class	AI fees per task
Ministral 3 3B	Workhorse class	0.0001 €
Qwen3.8 Flash (Together)	Workhorse class	0.0001 €
gpt-6-luna	Compact class	0.0001 €
Mistral Small 4	Workhorse class	0.0002 €
Ministral 3 8B	Workhorse class	0.0002 €
Cheapest model per class
Mistral Large 3	Frontier class	0.0007 €
Ministral 3 3B	Workhorse class	0.0001 €
gpt-6-luna	Compact class	0.0001 €

Extracting data from documents

Input 15,000 tokens, output 600 tokens, one call per task

Extracting data from documents
Model	Class	AI fees per task
Qwen3.8 Flash (Together)	Workhorse class	0.0013 €
Ministral 3 3B	Workhorse class	0.0014 €
gpt-6-luna	Compact class	0.0016 €
DeepSeek V4 Flash 0731 (Together)	Workhorse class	0.0020 €
Ministral 3 8B	Workhorse class	0.0021 €
Cheapest model per class
Mistral Large 3	Frontier class	0.0074 €
Qwen3.8 Flash (Together)	Workhorse class	0.0013 €
gpt-6-luna	Compact class	0.0016 €

Knowledge-base question

Input 8,000 tokens, output 400 tokens, one call per task

Knowledge-base question
Model	Class	AI fees per task
Ministral 3 3B	Workhorse class	0.0007 €
Qwen3.8 Flash (Together)	Workhorse class	0.0007 €
gpt-6-luna	Compact class	0.0009 €
Ministral 3 8B	Workhorse class	0.0011 €
DeepSeek V4 Flash 0731 (Together)	Workhorse class	0.0011 €
Cheapest model per class
Mistral Large 3	Frontier class	0.0041 €
Ministral 3 3B	Workhorse class	0.0007 €
gpt-6-luna	Compact class	0.0009 €

Research agent over documents

Input 25,000 tokens, output 1,200 tokens, 8 calls per task

Research agent over documents
Model	Class	AI fees per task
Qwen3.8 Flash (Together)	Workhorse class	0.0182 €
Ministral 3 3B	Workhorse class	0.0185 €
gpt-6-luna	Compact class	0.0218 €
DeepSeek V4 Flash 0731 (Together)	Workhorse class	0.0270 €
Ministral 3 8B	Workhorse class	0.0277 €
Cheapest model per class
Mistral Large 3	Frontier class	0.1007 €
Qwen3.8 Flash (Together)	Workhorse class	0.0182 €
gpt-6-luna	Compact class	0.0218 €

Writing or changing code

Input 15,000 tokens, output 1,500 tokens, 3 calls per task

Writing or changing code
Model	Class	AI fees per task
Ministral 3 3B	Workhorse class	0.0044 €
Qwen3.8 Flash (Together)	Workhorse class	0.0047 €
gpt-6-luna	Compact class	0.0059 €
Ministral 3 8B	Workhorse class	0.0065 €
DeepSeek V4 Flash 0731 (Together)	Workhorse class	0.0067 €
Cheapest model per class
Mistral Large 3	Frontier class	0.0258 €
Ministral 3 3B	Workhorse class	0.0044 €
gpt-6-luna	Compact class	0.0059 €

Working through a long document

Input 300,000 tokens, output 3,000 tokens, one call per task

Working through a long document
Model	Class	AI fees per task
Qwen3.8 Flash (Together)	Workhorse class	0.0245 €
DeepSeek V4 Flash 0731 (Together)	Workhorse class	0.0377 €
GLM-5.3-Flash (Together)	Workhorse class	0.0410 €
gpt-6-luna	Compact class	0.0548 €
gpt-5.4-nano	Workhorse class	0.0561 €
Cheapest model per class
DeepSeek V4 Pro	Frontier class	0.3592 €
Qwen3.8 Flash (Together)	Workhorse class	0.0245 €
gpt-6-luna	Compact class	0.0548 €

Meeting notes

Input 15,000 tokens, output 800 tokens, one call per task

Meeting notes
Model	Class	AI fees per task
Ministral 3 3B	Workhorse class	0.0014 €
Qwen3.8 Flash (Together)	Workhorse class	0.0014 €
gpt-6-luna	Compact class	0.0017 €
DeepSeek V4 Flash 0731 (Together)	Workhorse class	0.0020 €
Ministral 3 8B	Workhorse class	0.0021 €
Cheapest model per class
Mistral Large 3	Frontier class	0.0077 €
Ministral 3 3B	Workhorse class	0.0014 €
gpt-6-luna	Compact class	0.0017 €

Translation

Input 2,000 tokens, output 2,000 tokens, one call per task

Translation
Model	Class	AI fees per task
Ministral 3 3B	Workhorse class	0.0004 €
Ministral 3 8B	Workhorse class	0.0005 €
Ministral 3 14B	Workhorse class	0.0007 €
DeepSeek V4 Flash 0731 (Together)	Workhorse class	0.0007 €
Qwen3.8 Flash (Together)	Workhorse class	0.0007 €
Cheapest model per class
Mistral Large 3	Frontier class	0.0035 €
Ministral 3 3B	Workhorse class	0.0004 €
gpt-6-luna	Compact class	0.0011 €

What the tables leave out

The tables show AI fees only. Review time, rework, other running costs and the cost of errors depend on your own process, so they are not included. Since version 10 the AI cost calculator shows them separately, with your own figures. Whether a model is good enough for a scenario is best settled by a test on your own tasks.

Method

Prices are vendor list prices in USD per 1 million tokens from the tariff data of the AI cost calculator, oldest price check on 28 September 2026. Converted at the ECB reference rate of 29 September 2026: 1.1355 USD per euro.
AI fees per task: engine of the AI cost calculator, version 10, the scenario's "typical" size, one attempt, no cache, standard tier without batch processing, global processing. Reference date is 30 September 2026.
Offers whose context window or output limit cannot hold the task are left out of that scenario.
Median and factor (most expensive divided by cheapest model) are computed from the amounts of all models in the scenario, rounded to four decimal places.
Promotional and introductory prices count as long as they apply on the reference date. They are listed with their dates under "Tariff notes", as are announced shutdowns. Where a model concerned appears in a table, it carries the note there as well.
Included are all offers the calculator lists as current with a valid tariff on the reference date, including offers through hosts. The same model from two vendors or hosts counts as two offers.
The class comes from the AI cost calculator. It follows the tiers vendors use to name their models and states what a model is built for.
Your own sizes, cache shares, batch processing, regions and the costs beyond AI fees can be set in the calculator (German, version 10).

Tariff notes

gpt-5.6-sol: Promotional price until 21 November 2026
Gemini 3.8 Flash: Introductory rate until 31 December 2026
Gemini 3.7 Flash: Introductory rate until 31 December 2026
Gemini 3.6 Flash: Introductory rate until 31 December 2026
Gemini 3.1 Flash-Lite: Shutdown announced for 7 May 2027

Changes

30 September 2026: new edition based on version 10 of the AI cost calculator with 46 current models. The measure is AI fees per task. The section with solve rates has been dropped because the calculator has worked without solve rates since version 10.
10 September 2026: edition with 42 current models based on version 3.1 of the calculator, with machine cost per task and cost per solved task.

How to cite

Convios GmbH (2026): AI Cost Benchmark 2026, as of 30 September 2026, https://www.convios.com/en/insights/ki-kosten-benchmark-2026. Free to use with attribution (CC BY 4.0).

AI cost calculator, version 10 (German)Cutting AI cost without losing quality