Vana vs Jev: a cheaper Jev alternative
If you use Jev or are evaluating it, Vana is a System One model with the same request shape and a lower price per token. Compared here on what each vendor publishes, with an honest note on where Jev is the better fit.
The Vana team

Vana is a Jev alternative. Both are System One models: you send a state and some typed questions, and you get typed answers with probabilities back, not generated text. This post compares them using only what each vendor publishes. We have not run a new speed or accuracy benchmark for it.
What each one is
Vana 1.0 is our model, from Annamal AI (Mindsbeam Technologies). It reads English text, sent as a string or JSON, answers every question in one pass and gives each answer a confidence, so your code can decide when to act and when to hand off to a person.
Jev is TypeSafe AI's model. TypeSafe calls it "the first System One model" and launched it on 15 September 2026. It is trained with a method TypeSafe calls Reinforcement Learning for Calibrated Decisions (RLCD). The current releases are Jev 1.13 (jev-latest) and Jev 1.14 (jev-preview).
The API: nearly the same request
Both use POST /v1/systemone with a bearer key, a model, a state and a map of named questions (see the Vana reference and the Jev reference). Both support the same three question types: choice (pick one labelled option), noul (yes or no, as a probability) and score (a level on an ordered scale). Answers come back under the ids you used. Moving a simple integration is mostly a matter of changing the base URL, the key and the model name, but check which state shapes and confidence fields each one returns before you switch.
Pricing
Both bill on input tokens only. Vana's list price is about 17% lower per token.
| Vana 1.0 | Jev (1.13 / 1.14) | |
|---|---|---|
| Input, per 1M tokens | $0.035 | $0.042 |
| Input, per 1B tokens | $35 | $42 |
| Output tokens | Free | Free |
| 10B input tokens / month | $350 | $420 |
| 100B input tokens / month | $3,500 | $4,200 |
| Billing | Prepaid, top-ups from $5, no subscription | Per input token |
At 10B tokens a month: $350 vs $420. At 100B: $3,500 vs $4,200.
Monthly figures are list price times volume. What you actually pay also depends on how many tokens each model counts for the same text, so price your own inputs. Sources: Vana pricing, TypeSafe models page.
Free credits
| Vana 1.0 | Jev (1.13 / 1.14) | |
|---|---|---|
| Free credits on signup | 1M input tokens, one-time, after email verification, no card | None for new signups |
Vana gives every new account 1M free input tokens once, one per person, so you can try it on real traffic before you pay (see the credit terms). Jev offers no free credits to new signups: TypeSafe said on 27 September 2026 that "new signups no longer get free credits; we hope to bring these back as soon as possible". Its docs don't describe a free tier.
Limits
| Vana 1.0 | Jev (1.13 / 1.14) | |
|---|---|---|
| Max input (context) | 1,024 tokens per request | 64k per request, 32k per question; 262k opt-in |
| Questions per request | Up to 64 | No separate limit; must fit the context |
| Rate limits | 300 requests/min per key by default after first top-up; more on request | Set per organization (tokens/s and requests/s), shown in the console |
| Language | English | English primary; others handled, not equally well |
Jev reads far longer inputs. Vana is built for short states such as a message, a ticket or a short record.
Speed (vendor-stated)
| Vana 1.0 | Jev (1.13 / 1.14) | |
|---|---|---|
| Vendor-stated time | About 200 ms per call | 70–500 ms |
| How it is measured | Server-side, per call, any number of questions | End-to-end, from laptops on the US West Coast (launch post) |
These are each vendor's own figures, measured differently, so they can't be compared directly. If latency matters, measure both from where your code runs.
SDKs
| Vana 1.0 | Jev (1.13 / 1.14) | |
|---|---|---|
| Python | annamal-vana (pip install annamal-vana) | typesafe-sdk (pip install typesafe-sdk) |
| TypeScript / JavaScript | @annamal/vana (npm install @annamal/vana) | @typesafe-ai/sdk (npm install @typesafe-ai/sdk) |
| Other routes | HTTP API | Pydantic AI; gateways such as OpenRouter |
Vana has official Python and TypeScript SDKs with typed questions, retries with backoff on 429 and 5xx, timeouts and typed errors. See the Vana SDK docs. Jev's SDKs are described in TypeSafe's SDK docs.
Who builds each one
| Vana 1.0 | Jev (1.13 / 1.14) | |
|---|---|---|
| Launched | 9 Oct 2026 | 15 Sep 2026 (launch post) |
| Funding | Bootstrapped (self-funded, no outside investors) | Venture-funded |
| Latest round | None | $870M Series A, 9 Oct 2026 |
| Valuation | Not applicable | $7.5B |
| Lead investor | None | Andreessen Horowitz, with Sequoia Capital, DCVC and angels |
Annamal AI is bootstrapped. That keeps us focused on one thing, a fast decision model that does its job well, and it keeps our pricing disciplined and simple. We answer to our customers and decide our own roadmap. TypeSafe is venture-funded (TypeSafe's announcement, TechCrunch), which gives it room to build enterprise features and integrations quickly.
Where Jev may be the better fit
We'd rather you pick the right tool. Jev is probably the better choice when:
- Your inputs are long. For contracts, transcripts or long records, Jev's larger context wins; Vana truncates past its limit.
- You want it inside tools you already use. Jev is available through Pydantic AI and gateways such as OpenRouter.
- You need enterprise terms. TypeSafe offers custom and enterprise plans and describes zero data retention for enterprise customers.
Where Vana fits
Vana is built for short, high-volume decisions: routing a support ticket, flagging churn risk, scoring urgency, moderating a message or qualifying a lead. For that kind of work it is a cheaper Jev alternative with the same request shape, and you can try it free first.
At a glance
| Vana 1.0 | Jev (1.13 / 1.14) | |
|---|---|---|
| Vendor | Annamal AI (Mindsbeam) | TypeSafe AI |
| Launched | 9 Oct 2026 | 15 Sep 2026 |
| Endpoint | POST /v1/systemone | POST /v1/systemone |
| Question types | choice, noul, score | choice, noul, score |
| Input price per 1M / 1B | $0.035 / $35 | $0.042 / $42 |
| Output price | Free | Free |
| Free credits | 1M input tokens, one-time | None for new signups |
| Max input | 1,024 tokens, English | 64k per request (262k opt-in) |
| Vendor-stated speed | ~200 ms server-side | 70–500 ms end-to-end |
| SDKs | Python, TypeScript | Python, TypeScript |
| Funding | Bootstrapped | Venture-funded ($870M Series A) |
| Access | Self-serve at vana.annamal.ai | TypeSafe console, SDKs, gateways |
Jev details come from TypeSafe's public docs, launch post and post on X, and funding details from TypeSafe's Series A announcement and TechCrunch, as of 16 October 2026. Vendors change prices and limits, so check the source pages before you decide.
Try it
Make your first decision call.
Sign up, get 1M free input tokens and send your first request. No card needed.