Jev 1.14.0
The latest Jev preview model, adding experimental image input, Box localization, and an experimental 262K context mode.
Jev 1.14.0 is a hosted System One decision model available through jev-preview. It handles text plus danger-zone experimental images and adds the Box primitive for locating one visual target.
- Provider
- TypeSafe AI
- Reviewed
- October 19, 2026
- Project status
- Official
- Verification
- Verified
- Categories
- automation, development
- Compatibility
- TypeSafe API, Structured Outputs, Experimental Images, Box
Inputs
- ✓Text supplied as a string, JSON object, or array of text
- ✓Experimental images enabled with TypeSafe-Danger-Zone-Images: true
- ✓Typed decision questions and criteria
Outputs
- ✓Typed choice, score, ranking, classification, and extraction decisions
- ✓Box coordinates and confidence for one requested visual target
- ✓Confidence values
Interface
- ✓structured-output
Decision tasks
- ✓Classify
- ✓Choose
- ✓Score
- ✓Rank
- ✓Route
- ✓Verify
- ✓Extract
Verified model facts
Provider facts and directory notes are separated. Time-sensitive claims include attribution and should be checked again before production use.
- Current verified version
- jev-1.14.0
- Version aliases
- jev-preview
- Model family
- System One / Jev
- Release date
- 2026-10-14
- Access methods
- hosted-api, provider-sdk, rest-api
- Hosted API
- Yes
- Local deployment
- No
- Downloadable weights
- No
- License
- Unknown
- Supported languages
- English is the primary training language and currently strongest, Other languages, including CJK, are accepted but performance is uneven and requires workload-specific testing
Commercial-use note
Hosted commercial access is available under TypeSafe AI terms. No public weights or local deployment path are documented; verify current account terms before production use.
Context or input limits
The default limit is 64K tokens per request across state and all questions, with a separate 32K limit across state and the single longest question. Sending TypeSafe-Danger-Zone-Long-Context: true raises both per-request and per-question limits to 262K. Image input is Dreamer-ready and experimental, requires TypeSafe-Danger-Zone-Images: true, and may change.
Pricing
$42 per billion input tokens, equivalent to $0.042 per million input tokens; outputs are free according to the model documentation verified on October 19, 2026.
Latency and rate limits
Rate limits are set per organization and shown in the developer console. Requests over the configured quota return HTTP 429; TypeSafe SDKs retry by default, and teams can contact TypeSafe sales for higher limits.
Strengths
- ✓Adds Box localization for one requested visual target.
- ✓Supports experimental image input alongside typed decision primitives.
- ✓TypeSafe highlights multilingual improvement in the release notes.
Limitations
- ✓Image input and 262K long context are experimental and may change.
- ✓Box always returns a box, so target presence should be checked separately.
- ✓No local deployment or downloadable weights are documented.
- ✓jev-preview can move; jev-latest still resolves to Jev 1.13.0 as of October 19, 2026.
Safety notes
- ✓Use one target per Box question and pair localization with a presence check for consequential workflows.
- ✓Calibrate confidence thresholds on representative multilingual and image data.
- ✓Route uncertain or high-impact decisions to human review.
SDK and implementation links
Implementation guidance
Keep the declared inputs and outputs visible in code review. Validate every model response before it reaches a database, API, or customer-facing action.