Decision-model launches after Jev: October 2026 source tracker

the short answer

The October 2, 2026 snapshot has three kinds of new offerings. Perplexity and Cloudflare publish callable decision routes; OpenAI has announced a Decisions API but no verified public request contract. Intern-Decision, Julia 1, Laya and Kev-9B are independent model projects, while Ollama, Ollaya and SGLang provide serving paths. Check each source for weights, license, inputs and actual access. A similar question format gives you a starting fixture, not Jev-equivalent behavior.

Scope: Decision APIs, independent model projects and local serving routes
Verification: Primary documentation and public repositories checked October 2, 2026
Status rule: A model card, API contract and benchmark are different evidence
Next refresh: Recheck mutable access, versions, prices and aliases weekly during launch month

Hosted APIs and Provider Launches

The OpenAI row stops at an announcement. Cloudflare and Perplexity have published routes you can inspect and test. A model card tells you what artifact a publisher describes, but it cannot confirm your account has API access. For Jev version and provider changes, use the launch tracker.

Offering	Verified public surface	Boundary
TypeSafe Jev	Model and direct API documentation	Reference product; official weights are not public
OpenAI Decisions API	DevDay announcement	Public endpoint and probability schema not verified October 2
Cloudflare Clef and Clef-flash	Workers AI Clef route, launch post and model links	Cloudflare’s own models; reported benchmarks are vendor-run
Perplexity Decisions API	Published `/v1/decisions` reference, model card and pricing	Separate model and endpoint; test native behavior

Independent Models and Local Serving

Some repositories link to weights that may require a login or a separate approval. We recorded the artifact we could identify, not a download test for every checkpoint. Before committing to local serving, check the exact file, license and API behavior using the open-model guide. The table says nothing about which model wins on your task.

Offering	Primary artifact checked	Classification
Laya	Official model card and Python runtime	Independent local model family; multilingual variants
Kev-9B	Model card and source server	Independent open adapter and head
Intern-Decision	InternLM model cards, weights and inference source	Independent multimodal 0.8B, 2B and 4B family; benchmarks are publisher-run
Julia 1	Supersonic Labs model card, weights and ONNX/WebGPU artifact	Independent 144.3M model; Jev comparison uses supplied references
Vev	Source, checkpoint links and v0.1 evaluation	Image-capable research preview; weights are non-commercial
JevAlt	Source, three language-specific model links and evaluation data	Independent English, Turkish and German models; project-trained test sets favor them
Gutsy	Source, 0.8B GGUF model link and local server	Independent CPU-oriented model; performance figures are author-run
Mapika decider	Source, model links and release log	Independent open model family
Bespoke Nimble	Repository, model links and methodology	Independent model project; verify checkpoint access
Together Tev1	Training recipe and model link	Independent experimental model; verify checkpoint access
Firelex Jeff	Source, release notes and model links	Independent local model with task adapters
PostHog Jeeves	Source, evaluation and model links	Independent reasoning-plus-decision experiment
Ollama Nimble and Tev1	Official Nimble and Tev1 listings with local `/v1/systemone` examples	Ollama serving route for two separate model projects; requires Ollama 0.35+
Ollaya	Runtime source and model registry	Local serving runtime, not model weights
SGLang	Native `/v1/decisions` and `/v1/systemone` documentation	Serving framework for chat models; docs require a nightly build until release
openjev-sglang	Eric Zhang’s source and migration note	Earlier experiment; its README now directs users to SGLang’s native decisions API

How to Choose Which Launch Deserves a Separate Guide

A new repository does not automatically need its own comparison page. This dated list records what we found. The alternatives page is organized around the job you need done, and the named comparisons go deeper where a published contract or model card supports them.

Evidence and intent	Editorial action
Official API contract plus distinct Jev comparison query	Publish a dedicated comparison
Official model card plus local or language-specific decision	Publish a dedicated comparison with exact version
Repository and model link without stable public contract	Keep a sourced row and verify artifacts before expanding
Social launch without primary artifact	Track as a lead; do not state availability or benchmark claims

How to Keep This Tracker Accurate

Before calling one launch better for a particular workflow, run a paired benchmark on that decision. Vendor tables help you choose fixtures. The Jev limitations page lists failure cases to include when the incumbent is Jev.

Open each primary source and record the exact product, owner, version and verification date.
Check whether the endpoint or artifact is actually public; separate claims from tested access.
Re-run the same GEO prompt set for named alternatives and record factual errors in citations.
Recheck mutable pricing, limits and model aliases before editing evergreen comparisons.
Publish corrections beside the changed entry and prune pages with overlapping search intent.

FAQ

Did OpenAI release a Jev model?

No. OpenAI announced its own Decisions API. The public contract checked October 2, 2026 did not establish Jev compatibility or a shared model.

Which new decision APIs have published routes?

Perplexity documents /v1/decisions; Cloudflare documents Clef on Workers AI. Check live provider docs for current access and price.

Are the open models equivalent to Jev?

No equivalence follows from similar question names or published weights. Check conformance, target-task errors and probability calibration separately.

Sources

Checked against the sources below on October 2, 2026. Model versions, prices and limits change.

Change note: First source-checked ecosystem snapshot after the OpenAI, Cloudflare and Perplexity announcements.