AI Model Waddle

Version Reference

TypeSafe Jev Version & Model Changelog: Aliases, Releases, and Ecosystem Updates

Track official TypeSafe Jev model versions, alias targets (jev-latest, jev-preview), SDK updates, and gateway release milestones.

Published
Updated
  • Public fact

This page logs dated model releases, active model aliases, SDK versions, gateway availability, and limit changes for TypeSafe Jev.

As of September 2026, the current production model version is jev-1.13.0. The default SDK alias jev-latest routes to jev-1.13.0. For testing upcoming model iterations before promotion to stable, TypeSafe maintains the jev-preview alias.

Current Active Aliases and Endpoints

Alias / IdentifierCurrent TargetRecommended UsageProvider / Endpoint
jev-latestjev-1.13.0Default SDK target for production workloads; automatically tracks the latest stable checkpoint.POST api.typesafe.ai/v1/systemone
jev-1.13.0jev-1.13.0Explicitly pinned production version; guarantees identical weight representations and threshold calibration.POST api.typesafe.ai/v1/systemone
jev-previewjev-1.13.0 (currently identical)Pre-release alias designated for testing new checkpoints before general promotion.POST api.typesafe.ai/v1/systemone
typesafe/jev-1.13jev-1.13.0OpenRouter decisions alpha endpoint.POST openrouter.ai/api/alpha/decisions
typesafe-ai/jevjev-1.13.0Vercel AI SDK community provider identifier.Vercel AI SDK transport connector

Chronological Release Ledger

September 18, 2026: Gateway Expansions

  • OpenRouter Decisions Alpha: OpenRouter deployed a dedicated decisions API endpoint (POST /api/alpha/decisions) supporting typesafe/jev-1.13 with multi-provider routing and token billing.
  • Vercel AI SDK Community Provider: Community transport provider released enabling Jev integrations via standard @ai-sdk/provider interfaces.

September 15, 2026: TypeSafe Public Launch & Jev 1.13

TypeSafe AI emerged from stealth with a $40 million seed funding round led by DCVC, introducing the Jev System One decision model.

  • Production Model: jev-1.13.0 deployed to production inference cluster.
  • API Availability: POST https://api.typesafe.ai/v1/systemone opened in early-access.
  • Pricing Introduced: Input tokens priced at $0.042 per million ($42 per billion tokens); output tokens unmetered ($0.00 rate).
  • Core Primitives: Support for Choice (2–255 options), Score (2–10 ordered levels), and Noul (calibrated probability).
  • Early-Access Limits: Concurrency ceiling set to 1,200 requests per minute and 250,000 tokens per second per organization. Context window established at 64,000 tokens total (32,000 tokens for state plus longest question).
  • SDK Packages Released:
    • Python: typesafe-sdk published to PyPI.
    • TypeScript / JavaScript: @typesafe-ai/sdk published to npm.

Pinned Versions vs. Aliases: Best Practices

When deploying Jev to production software, choose between pinned versions and dynamic aliases based on your thresholding strategy:

  • Use pinned jev-1.13.0 if you use hardcoded confidence gates: Probability calibration curves can shift slightly between minor model releases. If your application code branches on strict numerical cutoffs (e.g., if confidence >= 0.91:), pin the explicit version jev-1.13.0 to avoid unintended automation behavior when new models are promoted.
  • Use jev-latest if you want automatic improvements: If your application uses Jev primarily for relative scoring, rankings, or coarse routing with human fallback queues, jev-latest ensures you automatically receive accuracy updates and latency improvements without redeploying code.

Related Context

Jev API Reference & Schema Specification

implementation

How to Use TypeSafe Jev: Setup and Quickstart

implementation

Jev Pricing & Workload Calculator

concept

Jev Rate Limits & Concurrency Ceilings

concept