On this site
Jev 1.13.0
TypeSafe · System One model
Leads the benchmark on calibrated probabilities and hard cases.
75.4JevBench v1.2 rank #1
Jev comparison · JevBench v1.2 · updated September 21, 2026
Jev vs djev
The fastest decision API, against the best calibrated one.
vs
Alternative
djev
Maisa · DiffusionGemma decision API
The fastest model on the board, and the only one that can look at a picture.
74.3JevBench v1.2 rank #3
- JevHosted API, generally availableWhere it runsdjevHosted API, still in free preview
- JevText onlySupported inputsdjevText, images, image options and live camera
- Jev0.65s medianResponse timedjev0.24s median
- JevCalibrated — 0.8 really does mean about 80%ProbabilitiesdjevMaisa documents them as experimental; do not threshold on them
- Jev74.1% correctHardest test casesdjev69.5% correct
- JevNoOpen sourcedjevThe community repo is Apache-2.0; the hosted service is not released
JevBench v1.2, measured the same way
One benchmark ran every system below under one method, so these bars are comparable with each other in a way that vendor-published figures are not.
Composite score
The top four finish within three points of each other. The composite hides where they actually differ, so the two charts below break it apart.
Capability, higher is better
IntelligenceHow often it picks the right answer
CalibrationWhether 0.8 really means about 80%
SpeedMeasured response time
Accuracy by how hard the decision is
Easy72 straightforward cases
Standard96 everyday cases
Judge146 evaluation-style calls
Hard220 genuinely ambiguous cases
Easy and standard decisions separate almost nothing. The hard tier, highlighted, is where these systems stop agreeing.
One thing these charts leave out: the composite score above also weighs a fourth axis, cost, scored from each system's public list price, and Jev places last of the five on it. We do not reproduce those figures, because Jev AI is an independent reseller metering its own credits and a list price is not what you would pay here. The full table is at the source below, and our rates are on the pricing page.
Every system on the board
The capability scores are charted above; this is the ranking and the deployment detail behind them. Jev and djev are highlighted.
| Model | Rank | Score | Measured latency | Runs on |
|---|---|---|---|---|
| Jev 1.13.0 (TypeSafe) | #1 | 75.4 | 0.65s median / 0.72s p95 | Hosted production API |
| SemIf (Qwen3.5-4B) | #2 | 74.7 | 0.20s raw, 0.55s adjusted | Self-hosted, RunPod GPU |
| djev (Maisa, DiffusionGemma) | #3 | 74.3 | 0.24s median / 0.31s p95 | Hosted API, free preview |
| openJev Verdict 1.4 | #4 | 72.5 | 0.31s raw, 0.78s adjusted | Self-hosted, CPU |
| Laya (ModernBERT-large) | #7 | 70.1 | 0.79s raw, 1.72s adjusted | Self-hosted, CPU |
JevBench v1.2 scored 42 systems on 534 decisions (72 easy, 96 standard, 146 judge, 220 hard) on 21 September 2026. The composite is a geometric mean of four equally weighted axes: intelligence, calibration, speed and cost. Self-hosted latency is adjusted by x2 plus 0.15s to approximate production conditions, so self-hosted numbers are not directly comparable with the raw latency their authors publish.
Jev and djev feature by feature
| Attribute | Jev 1.13.0 | djev |
|---|---|---|
| What it is | Hosted System One model from TypeSafe, version jev-1.13.0 | Hosted DiffusionGemma decision API from Maisa, in free preview |
| Question types | Noul (yes/no), choice, score — many per call, answered in parallel | Noul, choice, score |
| Inputs | Text only: string, JSON object or array | Text, native images, image options, live camera frames |
| Context | 64k tokens per request (32k for state plus the longest question) | Not published for the hosted preview |
| Calibration | Trained with RLCD; JevBench calibration 82.7 | Documented by Maisa as experimental and uncalibrated; JevBench 65.4 |
| Cost to run | Metered in Jev AI credits — see the pricing page | $0.026 per 1,000 decisions announced — nothing charged during preview |
| Open weights | No | Community djev-dev repo is Apache-2.0; Maisa’s hosted service is not yet released |
| Self-hosting | Not available | Possible: roughly one NVIDIA B200, Linux, CUDA 13, BF16 |
What each one is better at
Where Jev wins
- Calibration 82.7 against 65.4. Jev’s probabilities are trained and measured to mean something; djev’s are published as experimental.
- Hard-tier accuracy 74.1% against 69.5% — the gap opens exactly where the ambiguous cases live.
- A real price and real rate limits (250,000 tokens per second, 1,200 requests per minute) instead of preview terms that can change.
- A pinnable version. jev-latest resolves to jev-1.13.0 and you can name the version explicitly.
Where djev wins
- Fastest model in JevBench v1.2 — speed 91.4, 0.24s median, 0.31s p95.
- Reads images, scores image options and can sample a live camera. Jev cannot do any of that.
- Free while the preview lasts, so evaluation costs nothing but your time.
- Intelligence 88.4 is close enough to Jev that on easy and standard decisions you will rarely see a difference — both hit 100% easy and 97.9–99% standard.
The analysis
Where Jev and djev really differ, and why the bars above land where they do.
djev is genuinely fast. In JevBench v1.2 it takes the speed axis outright at 91.4 against Jev’s 83.3, with a 0.24s median response, and its intelligence score of 88.4 is within touching distance of Jev’s 90.4. If your bottleneck is milliseconds, djev is the credible answer.
The gap is calibration: 65.4 against Jev’s 82.7, and djev’s own documentation describes its probabilities as experimental and uncalibrated. That matters as soon as your code does anything other than take the top label — auto-approving above 0.9, escalating below 0.6, weighting a score. On the 220 hard decisions Jev is right 74.1% of the time against djev’s 69.5%.
The second gap is commercial. djev is a free preview: nothing is charged yet, $0.026 per 1,000 decisions is announced rather than billed, and open-sourcing is planned but has not happened. Jev is generally available, with published rate limits and a version you can pin.
What djev actually is
djev, the DiffusionGemma decision API from Maisa, by Maisa.
djev is built on Google’s DiffusionGemma running under vLLM. A normal language model writes an answer one token at a time; a diffusion model starts from noise across the whole output and refines it. djev exploits that by laying out a small “answer canvas” with one slot per question, denoising the slots together, and reading the probability of each allowed label straight off the result. There is no prose to parse and no JSON to repair.
Maisa runs the hosted service at api.djev.dev in free preview. A separate community repository, djev-dev, publishes the same approach under Apache-2.0 for self-hosting; it recommends a single NVIDIA B200 on Linux with CUDA 13 and unquantised BF16 weights, which is a serious machine. The repository reports sub-100ms warm text responses from internal measurement while noting those figures are not benchmarks for the current BF16 release and do not apply to image requests.
The distinctive capability is multimodal. djev accepts native image inputs, lets you use images as the options in a choice question, and can sample a live camera. Jev is text only: a string, a JSON object or an array. If your decision is about a photograph, Jev cannot make it and djev can.
So which should you use?
Pick Jev if
- Your code branches on the number, not just the argmax — auto-approve, escalate, weight, or route by confidence.
- Your inputs include long, messy or genuinely ambiguous cases rather than clean short ones.
- You need to forecast cost and hold a version steady across a release.
Pick djev if
- Your decision is about an image, a frame or a scanned document you have not converted to text.
- Latency is the binding constraint and you only need the winning label, not a trustworthy probability.
- You are comfortable building against a preview whose pricing and terms are not final.
Try Jev on the cases you are actually arguing about
If you want to see the calibration difference rather than read about it, run both on your own awkward cases. The Jev AI playground gives you Jev 1.13.0 with five free runs and no setup; every answer comes back with its full probability distribution and a confidence value, which is the thing you are actually comparing.
Jev vs djev FAQ
Is djev open source?
Partly. The community djev-dev repository is Apache-2.0 and shows how to run DiffusionGemma decisions on vLLM. Maisa’s hosted djev service says open-sourcing is planned but has not released its implementation.
Is djev free?
It is free during the preview — Benchmark Heaven notes that nothing is charged yet. Maisa has announced $0.026 per 1,000 decisions for afterwards, but that price is not being billed and the preview terms can still change. Jev AI’s own pricing is on the pricing page.
Can Jev read images like djev?
No. Jev accepts text only: a string, a JSON object or an array. Convert images to text or structured fields first, or use a model that takes pixels.
Which is faster, Jev or djev?
djev, on the hosted APIs JevBench measured: 0.24s median against Jev’s 0.65s. Both are fast enough for request-time use; neither is fast enough to sit inside a tight loop without batching.
Can I use djev probabilities as thresholds?
Maisa’s documentation calls them experimental and uncalibrated, so treat them as a ranking signal rather than a probability until you have validated them on your own labelled data.
Other Jev comparisons
Every number on these pages is quoted from a published source and was read on September 21, 2026.
- Jev vs SemIfThe strongest open rebuild on the board, and where it still falls short.
- Jev vs openJevNear-perfect on the data it was trained for, and shaky on anything else.
- Jev vs LayaA model you fine-tune on your data, against one that works out of the box.
Sources
Jev AI is an independent service and is not affiliated with or endorsed by TypeSafe. Jev answers are probabilistic; review important decisions before acting on them. djev and every other product named on this page belongs to its respective owner and is not affiliated with Jev AI. Figures are quoted from the sources above as they read on September 21, 2026 and can change.