Pricing that costs less than the tokens it saves.
The free tier includes 1M context tokens to try it, once. Paid plans include a monthly allowance, metered on the tokens you send, the same number your savings report measures against.
For evaluating on a real workload.
- 1M context tokens to try it, one time
- Agent SDK cleanup, always free
- Quality guard and savings report
- Per-agent budgets
- Community support
For production apps and agents.
- 100M context tokens / month
- $0.18 per extra 1M tokens
- Email support
For high-volume agents and platforms.
- 350M context tokens / month
- $0.15 per extra 1M tokens
- Everything in Pro
- Priority support
For platforms and regulated teams.
- Volume pricing
- Self-hosted deployment
- Custom decision models
- SLA and dedicated support
What you'd save, model by model.
Your model bills every input token. Contextwaise removes the ones it never needed, so you pay less, plan included.
| Model | Without Contextwaise | With Contextwaise | You save |
|---|---|---|---|
Claude Opus 5.5 $4 / 1M input | $2,095 | $728 $685 model + $43 Pro | $1,367 65% less |
Claude Sonnet 5.5 $2 / 1M input | $1,047 | $386 $343 model + $43 Pro | $661 63% less |
GPT-6.1 Sol $2 / 1M input | $1,047 | $386 $343 model + $43 Pro | $661 63% less |
Gemini 3.1 Pro $2 / 1M input | $911 | $351 $308 model + $43 Pro | $560 61% less |
DeepSeek V4 Pro $0.66 / 1M input | $274 | $132 $89 model + $43 Pro | $142 52% less |
GPT-6 Luna $0.1 / 1M input | $55 | $62 $19 model + $43 Pro | Costs more than it saves on a model this cheap |
You pay Contextwaise for 254.5M tokens a month: only the large tool outputs the SDK sends for scoring. Cleanup runs on your machine and is free. Cheapest plan: Pro, $43 a month with overage.
Monthly input cost at list prices, cache reads and writes included. Agent figures come from the run you can replay in the demo: 5 tasks, 17 model calls on GPT-6.1 Sol. Output tokens are the same either way and left out. Your savings report measures your own numbers.
The agent cache split follows automatic prompt caching. Anthropic models need cache breakpoints set (explicit caching) to get it.
Every plan includes the quality guard, the savings report and per-agent budgets.
You pay for the context tokens you send, counted the way the savings report counts them. A long tool output and a short chat turn cost what they weigh.
The agent SDK cleans every tool result on your machine, for free. Only results still over budget after cleanup go to the API for scoring, and only those count. On our demo agent that was at most 31,759 tokens, about a quarter of the 124,803 the model would have read without Contextwaise.
You keep calling your LLM provider directly. Contextwaise never sits in that billing path.
Yearly Pro or Scale started by December 31, 2026 costs 30% less than monthly for its first year: $126 for Pro, $411.60 for Scale. Applied at checkout, then it renews at the yearly price.
Pay yearly and save 20%: $144 a year for Pro, $468 for Scale.
Remove one call, or unwrap your agent, and your stack works exactly as before.
Questions about pricing
What counts as a context token?
The tokens in the context items you send to the API, counted at about four characters per token. The query does not count. Tokens Contextwaise removes still count: deciding they can go is the work. With the agent SDK, only the tool outputs it sends for scoring count; resending a result on later steps is free.
What happens past my included tokens?
On Pro and Scale, extra tokens are billed per started million: $0.18 on Pro, $0.15 on Scale, charged when each billing month ends. A billing month starts on the day you subscribed. Past about 290M tokens a month, Scale costs less than Pro with extra tokens. Free is a one-time 1M tokens: once used, requests return a 429 until you upgrade. Your app keeps working meanwhile, sending its original context.
When does Pro pay for itself?
Illustrative: at the full Pro allowance of 100M tokens, if Contextwaise removes 70% of the context, 70M input tokens never reach your model. Pro pays for itself once your model costs more than about $0.21 per million input tokens. On a model at $2 per million, that month avoids about $140 of input spend, about $125 after the plan.
Do you charge for LLM tokens?
No. You keep paying your model provider directly. Contextwaise decides what context reaches the model; it never resells or marks up model usage.
How do I know it is worth the price?
The savings report shows tokens removed, the LLM cost avoided and the quality check verdicts on your own workload. At launch, your savings report compares the cost avoided with the plan price.
Do agent tool results count?
Only the ones the agent SDK sends to the API for relevance scoring, counted the same way. Results it handles locally (cleanup, small outputs) are free and never leave your machine.
Is the quality guard a paid add-on?
No. Every plan includes the quality guard and the savings report, because cutting context without checking what was kept is not something we sell.
Cut your LLM costs
without cutting quality.
One layer between retrieval and inference removes the context your model never needed, in milliseconds. Fewer input tokens, lower cost, a prompt cache that keeps hitting, and every removal logged with its reason.