Official rate checked · daily source watch
TypeSafe Jev API pricing
The direct Jev decision API bills input tokens; output is free. Stable and preview routes are separate, and account limits apply.
Input · per 1M tokens
$0.042
Output · per 1M tokens
$0
Official source verification needs review. Shown rates are last verified values, not a newly confirmed quote.
Direct TypeSafe versus Opper routing
The current official model rate card and home-page price card agree on $0.042 per million input tokens ($42 per billion). Output tokens are free. The canonical feed, API mirror and reviewed source payload keep the direct rate separate from gateway fees.
Opper's official gateway pricing says it passes provider rates through without a token markup. It separately charges 3% on Gateway credit purchases or 5.5% with Control Plane features. BYOK on Gateway is free to Opper; taxes and optional checks can add cost. These are platform fees, so they are not added to the TypeSafe model rate in the canonical feed.
- Gateway: Pay as you go; Control Plane: Pay as you go; Enterprise: Custom contract.
- The Control Plane fee replaces the Gateway fee; the percentages are not added together. Optional LLM checks are billed at a flat rate per check in addition to the Control Plane fee.
- Only successful model responses are charged; failed and fallback attempts are not billed. No minimum spend or lock-in; premium models require a card.
- Without a card on file, management access is capped at 60 requests/minute and 3 active management keys; these are not inference rate limits.
What can Jev do?
The System One API returns typed decisions rather than free-form text. The documented stable alias jev-latest points to jev-1.14.0; jev-preview points to jev-1.14.0. Pin a version when calibrated thresholds must remain stable. These aliases are API routes, not subscription entitlements.
- Default context: 64,000 tokens per request; 32,000 per question.
- Optional long context: 262,000 tokens for both limits, enabled per request with
TypeSafe-Danger-Zone-Long-Context: true. The rate per token is unchanged; TypeSafe warns of slower, less accurate results. - Limits are organization-specific tokens/second and requests/second, shown in the console. Exceeding either returns HTTP 429; higher limits require custom or enterprise arrangements.
- Direct billing is per input token, with free output and no long-context token surcharge. Minimum spend, credit expiry, taxes and account-specific commercial terms are not established by the public model rate card; confirm them with TypeSafe.
Pac-Man benchmark versus production costs
Opper's Jevman cost analysis reports 100 games per model and a Jev cost around $0.00593 per game on its recorded setup from 2026-10-07. That is a vendor-run workload measurement, not a fixed Jev price per decision. The game uses fallback moves for late answers; your input size, call count, latency target, route, and gateway fee may differ.
Labs status
The current Labs 49-task text suite cannot import Opper's Pac-Man result. Jev's decision schema and real-time action loop require a separate pinned harness, full call and bill records, comparable controls, and an accepted-decision grader.
If you want to benchmark your own open-weight endpoint against Jevman's public game contract, check RunPod GPU options. We may earn a commission at no extra cost to you. RunPod does not host TypeSafe's proprietary Jev weights by implication.
Maintenance: current TypeSafe price, API access, limits and billing, plus Opper fees, have separate daily source-evidence clocks. Changed or unavailable sources stay unresolved until reconciled. The historical benchmark has its own date and is not refreshed by a current price check.