01
Categorical decision
Choice
Pick one answer from a declared set and receive the probability distribution across every option.
billing · technical · sales
Jev calculator:
cost & latency vs LLMs
Compare Jev and LLM costs for classification, routing, and scoring using your workload.
Interactive cost model
Put your classifier workload on the scale.
Change volume, context, and model assumptions. Every figure updates instantly in your browser.
01 / WorkloadAdjust inputs
Tickets, messages, records, or agent decisions per day.
1K10M
Full baseline request input budget, including instructions and all criteria. Jev uses the separate factor below.
5032K
JSON, labels, rationale, and parsing overhead per request.
04K
Jev input tokens ÷ baseline input tokens. Default 1.5× is a planning assumption; use 1× if counts match, or enter your measured ratio.
0.110
Estimated Jev input: 750 tokens/request. Formula: ceil(baseline input × factor). AY Automate reported 360/153 (≈2.35×) and 952/828 (≈1.15×) in two routing tasks. These are task-specific billed-input ratios, not universal tokenizer constants. Include instructions and criteria in your input budget; avoid counting them twice.
Assumed end-to-end time per call (ms). Default is not measured; replace with your own comparable measurement.
10010K
Editable assumption, not a guaranteed minimum or percentile.
1500
Editable assumption, not an SLA or guaranteed upper bound.
7060K
Assumptions: 30 days/month. Jev is priced at $0.042 per million input tokens with no metered output-token charge. Jev defaults to a 70–500ms per-call planning range from the TypeSafe launch report. Replace this range with your own measurements; it is not a measured p50/p95 or a guarantee. Network, workload and service load affect latency. Request durations are not summed into elapsed hours saved: concurrency and critical paths are not modeled. Baseline prices last verified 2026-09-22; verify them again before relying on a figure.
02 / EstimateLive · browser only
Estimated monthly impact
Jev vs GPT-5.6 Terra
1.5B baseline / 2.3B Jev input tokens per month
Monthly savings
$4,705.50
Estimated avoided model spend
Cost reduction
98.0%
Compared with the selected baseline
Jev response time
70ms–500ms
Per request · editable assumption, not measured latency
Monthly model cost
USD / MONTHJev$94.50
GPT-5.6 Terra$4,800.00
Jev latency assumption
70ms–500ms
GPT-5.6 Terra
1,200ms
Explore a real log · compare an assumed route
Where could Jev fit in an agent workflow?
Repository review · 12 recorded model requests · 11 tool calls. Explore a candidate Jev route while keeping script calls and the final answer on the LLM.
Recorded model cost
$0.019945Scenario model estimate
$0.006570Proposed model calls
9 Jev / 3 LLMRecorded usage; hypothetical substitution. Jev suitability and task quality are unverified. This example assumes the same input count for selected requests, code-supplied arguments and unchanged downstream paths. Adjust budgets in the simulator; these are not measured savings.
From the JevCalc blog / Case study 01
When is LLM fallback worth it?
We tested Jev, Gemini, and a confidence-gated cascade on BANKING77. The development data led our updated selector to choose Jev-only.
Jev vs traditional LLMs
A decision engine, not a smaller chatbot.
Jev trades free-form generation for typed answers, predictable structures, and low-latency decisions that software can consume directly.
| Feature | Jev | Generative LLM |
|---|---|---|
| Output type | Typed Choice / Score / Noul | Generated text or JSON |
| Output token price | $0 | $0.40–$50.00 / 1M |
| Format reliability | 0% type-error rate by design | Requires validation |
| Per-call latency reference (not P50) | 70–500ms · vendor-reported | 1,200ms · editable assumption |
“0% type-error rate” describes schema validity, not perfect semantic accuracy. The 70–500ms Jev range is vendor-reported, not a measured percentile or a guarantee; the LLM value is a planning assumption. Benchmark comparable calls on your own workload.
Jev primitives cheat sheet
Three outputs. One shared state.
Ask multiple independent typed questions against the same context and evaluate them in parallel.
02
Position on a rubric
Score
Rate a state against two to ten ordered levels, with a continuous score and confidence attached.
low · medium · high
03
Yes / no probability
Noul
Return the probability that a statement is true without generating an explanation or extra text.
P(true) = 0.94
API limits & specifications
Plan the workload before it reaches production.
Use published limits as a starting point, then confirm current quotas and model availability in your TypeSafe account.
32K / 64K
Context guidance
32K state + longest question; 64K request
$0.042
Input price
Per one million input tokens
$42
Billion-token price
Equivalent published input rate
Typed
Answer contract
Choice, Score, or Noul
Frequently asked questions
What teams ask before switching.
How much does Jev cost?
The published rate used by this calculator is $0.042 per million input tokens, equivalent to $42 per billion input tokens. Output tokens are not separately metered.
Is Jev a replacement for GPT or Claude?
No. Jev returns bounded decisions and probabilities rather than free-form text. It is best used for classification, routing, scoring, and verification alongside a generative model.
What does zero percent hallucination mean for Jev?
It means Jev cannot produce a value outside the answer schema you define. It can still select a valid but incorrect option, so confidence thresholds and evaluation remain important.
Does this calculator send my workload data anywhere?
No. The MVP calculator runs entirely in your browser. Values are not submitted to JevCalc or TypeSafe AI.
Build on top of the decision layer
Need enterprise confidence routing, audit logs, and a fallback gateway?
Join the early-access list for production tooling around Jev and generative-model fallbacks. Sign up for launch and early-access updates; joining does not grant console access.