How much does it cost to summarize a document with AI?

Summarizing a 10-page document (≈5,000 tokens input, 500 tokens output) costs $0.00075 with Gemini Flash, $0.001 with DeepSeek V4 Flash, $0.003 with GPT-4o mini, and $0.015 with Claude Opus 4.8. For 1,000 documents/month, that's $0.75 to $15.

How much does AI code generation cost?

Generating a 200-line function (≈2,000 tokens input context, 1,000 tokens output) costs $0.0007 with Gemini Flash, $0.0013 with DeepSeek V4 Flash, $0.004 with GPT-4o, and $0.03 with Claude Opus 4.8. At 100 generations/day, monthly cost ranges from $2.10 to $90.

What is the cheapest AI API for chatbots?

Gemini 2.0 Flash ($0.10/$0.40 per 1M tokens) is cheapest for high-volume chatbots. A typical conversation (1K input, 500 output tokens) costs $0.0003. DeepSeek V4 Flash ($0.14/$0.28) is nearly as cheap with better code understanding. Both handle 10K conversations/day for under $15/month.

How much does AI data extraction cost?

Extracting structured data from a 1,000-word document (≈1,500 tokens input, 300 tokens output) costs $0.00023 with Gemini Flash, $0.0003 with DeepSeek V4 Flash, $0.002 with GPT-4o, and $0.01 with Claude Sonnet 4.6. Processing 10,000 documents/month costs $2.30 to $100.

← Back to blog

Analysis June 5, 2026

AI API Cost Per Task: What 10 Common Tasks Actually Cost in 2026

Most developers think in "cost per token." But what you really want to know is: how much does it cost to actually do the thing? Here are real costs for 10 common AI tasks — with exact token counts and dollar amounts across every major provider.

TL;DR: The cheapest provider for any task costs 10-50x less than the most expensive. For most tasks, Gemini Flash or DeepSeek V4 Flash costs fractions of a cent per operation. At scale, the difference between providers is thousands of dollars per month.

How to Read These Numbers

Each task below shows:

Typical token count — how many input/output tokens the task uses
Cost per operation — what one task costs on each provider
Monthly estimate — what it costs at common volume levels

Token counts are based on real-world usage patterns. 1 token ≈ 4 characters of English text.

1. Document Summarization

~5,000 input tokens → ~500 output tokens

Summarize a 10-page document (meeting notes, report, article) into 3-5 bullet points.

Model	Cost per Summary	1,000/month
Gemini 2.0 Flash	$0.00075	$0.75
DeepSeek V4 Flash	$0.00109	$1.09
GPT-4o mini	$0.00285	$2.85
Claude Sonnet 4.6	$0.0225	$22.50
Claude Opus 4.8	$0.0375	$37.50

Best choice: Gemini Flash at $0.75/month for 1K summaries. Claude Opus costs 50x more for the same task.

2. Code Generation

~2,000 input tokens → ~1,000 output tokens

Generate a 200-line function with comments, error handling, and types. Typical Copilot-style completion.

Model	Cost per Generation	100/day (3K/month)
Gemini 2.0 Flash	$0.0006	$1.80
DeepSeek V4 Flash	$0.00084	$2.52
GPT-4o	$0.0125	$37.50
Claude Sonnet 4.6	$0.021	$63.00
Claude Opus 4.8	$0.035	$105.00

Best choice: Gemini Flash or DeepSeek for autocomplete. For complex code review, GPT-4o at $37.50/month is the quality/cost sweet spot.

3. Chatbot Conversation Turn

~1,000 input tokens → ~500 output tokens

One exchange in a customer support or FAQ chatbot — user message + context, bot response.

Model	Cost per Turn	10K conversations/day
Gemini 2.0 Flash	$0.0003	$9.00
DeepSeek V4 Flash	$0.00028	$8.40
GPT-4o mini	$0.0009	$27.00
Claude Haiku 4.5	$0.0035	$105.00
Claude Sonnet 4.6	$0.0105	$315.00

Best choice: DeepSeek V4 Flash at $8.40/month for 10K conversations/day. Even at high volume, budget models keep costs under $30.

4. Structured Data Extraction

~1,500 input tokens → ~300 output tokens

Extract names, dates, amounts, or categories from unstructured text into JSON.

Model	Cost per Extraction	10K documents/month
Gemini 2.0 Flash	$0.00023	$2.25
DeepSeek V4 Flash	$0.00030	$3.00
GPT-4o mini	$0.00068	$6.75
Claude Sonnet 4.6	$0.009	$90.00
Claude Opus 4.8	$0.0128	$127.50

Best choice: Gemini Flash or DeepSeek for bulk extraction. At $2-3/month for 10K documents, even indie projects can afford structured extraction.

5. Email Drafting

~800 input tokens → ~400 output tokens

Draft a professional email from a brief prompt. Sales outreach, support reply, or internal update.

Model	Cost per Email	500/day (15K/month)
Gemini 2.0 Flash	$0.00024	$3.60
DeepSeek V4 Flash	$0.00027	$4.05
GPT-4o mini	$0.00072	$10.80
Claude Sonnet 4.6	$0.0084	$126.00
Claude Opus 4.8	$0.014	$210.00

Best choice: Any budget model handles email well. DeepSeek at $4.05/month for 500 emails/day is the sweet spot for quality vs cost.

6. Content Classification / Sentiment Analysis

~500 input tokens → ~50 output tokens

Classify support tickets, categorize feedback, or analyze sentiment. Short input, tiny output.

Model	Cost per Classification	50K items/month
Gemini 2.0 Flash Lite	$0.00005	$2.63
Gemini 2.0 Flash	$0.00007	$3.50
GPT-4o mini	$0.00038	$18.75
DeepSeek V4 Flash	$0.00021	$10.50
Claude Haiku 4.5	$0.0028	$137.50

Best choice: Gemini Flash Lite — the cheapest model at $0.075/M input. At $2.63/month for 50K classifications, this is almost free.

7. Translation (1 page)

~1,500 input tokens → ~1,500 output tokens

Translate a one-page document between languages. 1:1 token ratio for translation tasks.

Model	Cost per Page	1K pages/month
Gemini 2.0 Flash	$0.00075	$0.75
DeepSeek V4 Flash	$0.00063	$0.63
GPT-4o mini	$0.00315	$3.15
Mistral Small 4	$0.00293	$2.93
Claude Sonnet 4.6	$0.027	$27.00

Best choice: DeepSeek V4 Flash — symmetric pricing means input and output cost the same. $0.63/month for 1K pages.

8. RAG / Document Q&A

~4,000 input tokens (context + question) → ~500 output tokens

Answer a question using retrieved context from your knowledge base. Typical RAG pipeline output.

Model	Cost per Query	5K queries/day
Gemini 2.0 Flash	$0.0006	$90.00
DeepSeek V4 Flash	$0.0007	$105.00
GPT-4o	$0.015	$2,250
Claude Sonnet 4.6	$0.0195	$2,925
Claude Opus 4.8	$0.0325	$4,875

Best choice: RAG is token-heavy (lots of context). Budget models save 98% vs premium. Gemini Flash at $90/month vs Opus at $4,875/month — for the same queries.

9. Image Description / Alt Text Generation

~1,000 input tokens (image embedding) → ~200 output tokens

Generate descriptive alt text or captions for images. Multimodal input, short text output.

Model	Cost per Image	5K images/day
Gemini 2.0 Flash	$0.00018	$27.00
GPT-4o mini	$0.00055	$82.50
GPT-4o	$0.0045	$675.00
Claude Sonnet 4.6	$0.006	$900.00

Best choice: Gemini Flash is the cheapest multimodal model. Not all models support image input — check compatibility before choosing.

10. AI Agent / Multi-Step Reasoning

~3,000 input tokens → ~1,500 output tokens per step, ~5 steps

An AI agent that plans, executes, and iterates. Each "thought" is a full API call with context.

Model	Cost per Task	100 tasks/day
Gemini 2.0 Flash	$0.0045	$135.00
DeepSeek V4 Flash	$0.0042	$126.00
GPT-5 mini	$0.0188	$562.50
Claude Sonnet 4.6	$0.0675	$2,025
Claude Opus 4.8	$0.1125	$3,375

Best choice: Agents are the most expensive use case because they multiply token usage. Use budget models for simple steps, premium for complex reasoning.

The Cost Matrix: Quick Reference

Here's every task on every provider at a glance. Numbers show cost per single operation.

Task	Gemini Flash	DeepSeek V4F	GPT-4o mini	Sonnet 4.6	Opus 4.8
Document Summary	$0.00075	$0.00109	$0.00285	$0.0225	$0.0375
Code Generation	$0.0006	$0.00084	$0.003	$0.021	$0.035
Chatbot Turn	$0.0003	$0.00028	$0.0009	$0.0105	$0.0175
Data Extraction	$0.00023	$0.0003	$0.00068	$0.009	$0.0128
Email Drafting	$0.00024	$0.00027	$0.00072	$0.0084	$0.014
Classification	$0.00007	$0.00021	$0.00038	$0.0023	$0.0038
Translation	$0.00075	$0.00063	$0.00315	$0.027	$0.045
RAG Q&A	$0.0006	$0.0007	$0.0045	$0.0195	$0.0325
Image Description	$0.00018	—	$0.00055	$0.006	$0.01
Agent (5 steps)	$0.0045	$0.0042	$0.0188	$0.0675	$0.1125

Key Takeaways

Budget models are 10-50x cheaper than premium models for most tasks. Use Gemini Flash or DeepSeek V4 Flash for high-volume, routine work.
Agents are the most expensive use case. Each "step" is a full API call — multiply your per-call cost by the number of reasoning steps.
RAG is token-heavy. Sending 4K tokens of context per query adds up fast. Consider caching frequent queries.
The cheapest provider changes by task. Gemini Flash wins on summarization, DeepSeek wins on chat and translation. There's no single "cheapest" model.
Premium models are worth it for complex reasoning. Claude Opus and GPT-5.5 shine on tasks that require deep understanding, not just pattern matching.

Calculate Your Exact Costs

Enter your token usage and see exactly what each provider charges. Cheapest options ranked automatically.

Open Cost Calculator → Model Status Dashboard Track Costs Over Time →