LangGraph cost calculator
LangGraph agents loop through nodes, call tools, retry edges, and hand off between sub-agents — each step re-sending a growing state. Model your graph's calls, loops, retries and handoffs to see cost per task, monthly spend, and where tokens leak.
Switching from Claude Sonnet 4.6 to DeepSeek-V4 Pro (reasoner) (DeepSeek, same frontier tier) cuts monthly spend by $3.5K — 86% — for this workload.
Where each task's cost goes
| Model | $/task | $/successful | Monthly | vs current |
|---|---|---|---|---|
CHEAPEST Groq · fast | $0.0357 | $0.0397 | $71.40 | cheapest |
| OpenAI · fast | $0.036 | $0.04 | $71.95 | cheapest |
| Gemini · fast | $0.0671 | $0.0746 | $134 | 1.9× |
| OpenRouter · mid | $0.0733 | $0.0815 | $147 | 2.1× |
| DeepSeek · mid | $0.0897 | $0.0996 | $179 | 2.5× |
| OpenRouter · mid | $0.1486 | $0.1651 | $297 | 4.2× |
| OpenAI · mid | $0.1799 | $0.1998 | $360 | 5.0× |
| Gemini · mid | $0.217 | $0.2412 | $434 | 6.1× |
| Together · mid | $0.2378 | $0.2642 | $476 | 6.7× |
| DeepSeek · frontier | $0.2783 | $0.3092 | $557 | 7.8× |
| Groq · mid | $0.4194 | $0.466 | $839 | 11.7× |
| Anthropic · mid | $0.6832 | $0.7591 | $1.4K | 19.1× |
| Together · mid | $0.735 | $0.8167 | $1.5K | 20.6× |
| OpenAI · frontier | $0.8993 | $0.9992 | $1.8K | 25.2× |
| Gemini · frontier | $0.8993 | $0.9992 | $1.8K | 25.2× |
| OpenAI · frontier | $1.37 | $1.52 | $2.7K | 38.3× |
CURRENT Anthropic · frontier | $2.05 | $2.28 | $4.1K | 57.4× |
| Anthropic · frontier | $3.42 | $3.80 | $6.8K | 95.7× |
Per-1M-token list prices; figures are estimates and exclude embeddings, fine-tuning, image/audio, and infrastructure. Rows marked verifystill need a final check against the provider's live pricing. LLM prices change often.
Frequently asked questions
Why do LangGraph agents get expensive?+
Cyclic graphs re-enter nodes and re-send the accumulated state each time. A loop that runs a few extra iterations, or a handoff that copies the full context to a sub-agent, multiplies token usage fast. The calculator's loop, handoff and context inputs make that cost visible.
How do I estimate LangGraph token costs before running?+
Estimate the average number of model calls per task, the tokens each tool node returns into state, and how much context accumulates per turn. Plug those in here; the model is the same accumulate-and-re-send loop LangGraph executes.
Do agent handoffs cost extra tokens?+
Yes. Each handoff re-reads the stable prefix plus the accumulated conversation in the receiving agent's context. Set the handoff count and watch per-task cost climb — it's one of the most underestimated LangGraph line items.
Which model is cheapest for a LangGraph agent?+
It depends on the graph. The comparison table runs your exact graph profile across every provider so you can see, for example, whether routing tool-heavy nodes to DeepSeek-V3 or Gemini Flash beats running everything on a frontier model.