AI Coding Agent Cost Calculator
Estimate what a coding task costs across models. Pick a task size. Add a retry multiplier, which most calculators leave out. Then compare the per-task cost of the models coding agents run on. All prices are API list rates checked against each vendor's source.
Estimated cost per task (30,000 in / 8,000 out, ×1.5)
| Model | Vendor | Est. cost / task |
|---|---|---|
| DeepSeek V4 (flash) | DeepSeek | $0.0097 |
| GPT-5.4 nano | OpenAI | $0.02 |
| GLM-4.6 | Z.ai | $0.05 |
| GPT-5.4 mini | OpenAI | $0.09 |
| Claude Haiku 4.5 | Anthropic | $0.11 |
| GLM-5.2 | Z.ai | $0.12 |
| Gemini 3.6 Flash | $0.16 | |
| Claude Sonnet 5 | Anthropic | $0.21 |
| Claude Opus 5 | Anthropic | $0.52 |
| GPT-5.6 (Sol) | OpenAI | $0.59 |
An estimate, not a quote. Token counts vary widely by task, codebase, and how the agent is driven. Prices are per-token API list rates, checked against each vendor's pricing page on the dates shown below. Provider discounts, caching, and batch rates can lower them.
Prices and sources
Per-1M-token API list rates. Last checked 2026-08-11. Models that price higher above ~200k tokens (Gemini Pro, Grok) are left out rather than shown at a misleading flat rate. Token counts also differ by model: recent Claude models count roughly 30% more tokens for the same text.
- Claude Opus 5$5/M in · $25/M outsource
- Claude Sonnet 5$2/M in · $10/M outsource
- Claude Haiku 4.5$1/M in · $5/M outsource
- GPT-5.6 (Sol)$5/M in · $30/M outsource
- GPT-5.4 mini$0.75/M in · $4.5/M outsource
- GPT-5.4 nano$0.2/M in · $1.25/M outsource
- Gemini 3.6 Flash$1.5/M in · $7.5/M outsource
- GLM-5.2$1.4/M in · $4.4/M outsource
- GLM-4.6$0.6/M in · $2.2/M outsource
- DeepSeek V4 (flash)$0.14/M in · $0.28/M outsource
Common questions
- How much does it cost to build with an AI coding agent?
- It depends on the model, the size of the task, and how many passes the agent takes. This calculator takes each model's per-token API price. It multiplies that by the tokens a task uses, then by a retry multiplier. That lets you compare models on the same task. The output is an estimate, not a quote.
- Why include a retry multiplier?
- Because real coding runs rarely land in one pass. An agent reads context, tries, tests, and often revises. Costing only the first try understates the real cost. So this tool lets you add a realistic multiplier.
- Are these the real prices?
- They are each vendor's published per-token API list rate. Each was checked against the vendor's own pricing page on the dates shown under 'Prices and sources'. List rates change. Discounts, prompt caching, and batch pricing can lower them. So treat the result as a planning estimate.
- How many tokens does a coding task use?
- There is no single number; it varies widely by task and codebase. The presets here are rough starting points, so adjust the input and output token fields to match your own work for a closer estimate.