💸 Cost-Per-Task Benchmarks
Forget cost per token. Here's what it actually costs to do real work.
Per-token pricing makes it nearly impossible to compare AI models at a glance — you'd need to estimate prompt length, output size, and context overhead just to get a rough number. So we've done that work for you. Below you'll find pre-calculated cost estimates for the most common real-world tasks across the leading text and image models. Green = cheapest, red = most expensive in each row.
Writing Tasks
| Task | GPT-4o | Claude 3.7 | Gemini 2.0 | Mistral Large |
|---|---|---|---|---|
| 500-word blog post | $0.008 | $0.009 | $0.005 | $0.004 |
| 1,500-word article | $0.024 | $0.027 | $0.015 | $0.012 |
| 10 social media posts | $0.003 | $0.003 | $0.002 | $0.001 |
| Product description (200 words) | $0.003 | $0.003 | $0.002 | $0.001 |
| Email sequence (5 emails) | $0.010 | $0.011 | $0.007 | $0.005 |
Research & Analysis Tasks
| Task | GPT-4o | Claude 3.7 | Gemini 2.0 | Mistral Large |
|---|---|---|---|---|
| Summarise a 10-page PDF | $0.012 | $0.014 | $0.008 | $0.006 |
| Competitive analysis (3 companies) | $0.035 | $0.040 | $0.022 | $0.018 |
| Extract key data from 50-row CSV | $0.008 | $0.009 | $0.005 | $0.004 |
| Literature review summary | $0.020 | $0.023 | $0.013 | $0.010 |
Coding Tasks
| Task | GPT-4o | Claude 3.7 | Gemini 2.0 | Mistral Large |
|---|---|---|---|---|
| Write a REST API endpoint | $0.015 | $0.017 | $0.010 | $0.008 |
| Debug a 100-line function | $0.010 | $0.012 | $0.007 | $0.005 |
| Write 10 unit tests | $0.012 | $0.014 | $0.008 | $0.006 |
| Refactor a component (200 lines) | $0.018 | $0.021 | $0.012 | $0.009 |
| Generate SQL query from plain English | $0.005 | $0.006 | $0.003 | $0.002 |
Image Generation Tasks
| Task | DALL-E 3 | Midjourney v6 | Stable Diffusion | Firefly |
|---|---|---|---|---|
| 1 hero image (1024×1024) | $0.040 | $0.033 | $0.002 | $0.029 |
| 10 product images | $0.400 | $0.330 | $0.020 | $0.290 |
| Social media graphic set (4 images) | $0.160 | $0.132 | $0.008 | $0.116 |
| Logo concept (5 variations) | $0.200 | $0.165 | $0.010 | $0.145 |
How we calculated this
Each cost is calculated as (input_tokens × input_price) + (output_tokens × output_price), using published API pricing as of Q1 2026.
Token assumptions:
- 500-word blog post: ~400 input tokens, ~700 output tokens
- 1,500-word article: ~600 input tokens, ~2,100 output tokens
- 10 social media posts: ~200 input tokens, ~400 output tokens
- 10-page PDF summary: ~7,000 input tokens, ~600 output tokens
- Competitive analysis: ~500 input, ~4,500 output tokens
- REST API endpoint: ~300 input, ~800 output tokens
- Component refactor (200 lines): ~1,200 input, ~1,200 output tokens
Image prices reflect per-image API costs: DALL-E 3 standard ($0.040/image), Midjourney v6 via fast GPU subscription (~$0.033/image), Stable Diffusion XL on cloud inference (~$0.002/image), and Adobe Firefly API (~$0.029/image).
All figures are estimates. Prices change frequently — always check the provider's current pricing page before making purchasing decisions.
Want to benchmark these models live with your own prompts?
Run real tasks, measure latency, and compare actual output quality — not just cost estimates.