[ Costs ]
What Jev actually costs
[ In short ]
The economics in four points
You pay for input only
Input only: the item plus the text of your questions. Output is free. Billing is with TypeSafe AI.
Under 5¢ per 1,000 items
Our list-price estimate for typical workloads. Reported runs range from 3¢ to 70¢ per 1,000, mostly because of longer inputs.
5–8× cheaper than small LLMs
One independent test against GPT-5.4 nano and Gemini 3.5 Flash-Lite on the same classification tasks.
Input length drives cost
A subject line costs far less than a full email thread. Send only what the decision needs.
[ Reported runs ]
What builders have paid
Public runs with the cost per 1,000 worked out. Self-reported figures are the author's own; we haven't reproduced them.
| Workload | Volume | Total cost | Per 1,000 | Time | Source |
|---|---|---|---|---|---|
| Email classificationRiley Brown | 500 emails | 3.5¢ | 7¢ | a few seconds | self-reportedMade with Jev (opens in a new tab) |
| Gmail triage (open source: jevmail)Syed Fazle Rahman, co-founder of HashnodeSorts into Needs reply, Updates, Promos, Sales and Spam via Vercel AI Gateway. | 1,000 emails | 3¢ | 3¢ | about 1 minute | self-reportedGitHub: fazlerocks/jevmail (opens in a new tab) |
| Lead scoringRomàn, indie software builderScored fit, confidence and mismatch before deciding which records deserved deeper review. | 700 leads | 9¢ | 13¢ | 40 seconds | self-reportedMade with Jev (opens in a new tab) |
| Classifying AI research papersHassan El Mghari, developer | 1,018 papers | 8¢ | 7.9¢ | 256 ms median latency | self-reportedMade with Jev (opens in a new tab) |
| Fraud email detectionHassan El Mghari, developerAuthor reports 96% accuracy on their test set. | 100 emails | 7¢ | 70¢ | 1.42 seconds | self-reportedMade with Jev (opens in a new tab) |
| Post virality scoring (61 questions per draft)Rob Hallam, maker of SuperX61 questions batched over one draft. | 1 draft | under 1¢ | — | about 1 second | self-reportedMade with Jev (opens in a new tab) |
| Browser agent: flight searchGregor Zunic, co-founder of Browser Use | 1 task | under 1¢ | — | 7 seconds | self-reportedMade with Jev (opens in a new tab) |
| Natural-language property searchJustine MooreEvaluated listings against a plain-English brief. | 1 search | 18¢ | — | under 20 seconds | self-reportedMade with Jev (opens in a new tab) |
| Minecraft gameplay agentRonak Malde | 1 session (8m 43s) | 97¢ | — | — | self-reportedMade with Jev (opens in a new tab) |
[ Compared ]
Jev vs small language models
Cost per 1,000 decisions on the same tasks, from an independent test of 791 labeled decisions on 19 September 2026 (AY Automate, measured test (opens in a new tab)).
| Task | Jev | GPT-5.4 nano | Gemini 3.5 Flash-Lite |
|---|---|---|---|
| 8-way routing | 1.5¢ | 7¢ (4.7×) | 8.7¢ (5.8×) |
| 77-way routing | 4¢ | 20¢ (5.1×) | 30¢ (7.5×) |
| Prompt-injection check | 1.4¢ | 8.1¢ (5.9×) | 8.6¢ (6.3×) |
Worth knowing: on the 8-way routing task Jev used 360 input tokens against 153 for the other models. It was still cheaper because its per-token price is lower and output is free, but budget for more tokens than you'd expect. This test compared cost, not accuracy.
Why costs vary 23×
Across the runs above, cost per 1,000 items ranges from 3¢ to 70¢. The price is the same for everyone, so the difference is in what gets sent:
- Input length. A subject line is about 20 tokens; a full email thread can be thousands. This is the biggest lever.
- Number of questions. Each question adds its instructions and option text to the bill.
- Batching. Five questions in one request read the input once; five requests read it five times.
- Option count. 77-way routing cost about 2.6× more than 8-way routing in the measured test.
- Free windows. Some runs happened during promotional periods, so reported totals may be list-price equivalents.
The takeaway: trim the input to what the decision needs, and batch your questions.
Price your own workload
Volume, item length and questions per item. It takes a minute.
Open the calculatorPublished a Jev benchmark?
Send us the link with volume, cost and time. We add runs that show their numbers. Get in touch.
[ For developers ]
Exact price and limits
As of 22 September 2026. Sources: TypeSafe API reference (opens in a new tab), AY Automate pricing analysis (opens in a new tab).
- Price
- $0.042 per 1M tokens
- Input only. A token is roughly ¾ of a word.
- Output
- Free
- Answers aren't billed.
- Per-request limit
- 64,000 tokens
- Input plus all questions combined.
- Input + longest question
- 32,000 tokens
- Keep inputs compact.
- Options per question
- Up to 255
- Each option adds tokens.
- Latency
- 70–500 ms
- Per request; varies with input size and provider.