# Tev1-4B model sizes and sources

> Tev1-4B is Together AI's Qwen3.5-4B decision experiment. It reads 2 to 24 lettered options and generates its answer as one letter.

Updated 2026-11-04 (Australia/Adelaide)

By Together AI · Released: 2026-09-23 · OPEN

Together's Jev-inspired experiment: a Qwen3.5-4B fine-tune that reads a state, a question and 2 to 24 lettered options. Together says plainly it is not a non-autoregressive runtime like [Jev](<https://decisionmodels.ai/models/jev>): it keeps a standard next-token head and generates its answer as one letter.

## Facts

- Base: Qwen3.5-4B
- Licence: Code MIT; weights licence being finalised as of 2 Oct 2026
- Run it: Together serverless, or Ollama

## Claims and sources

- **VENDOR-REPORTED**: Together reports 880/1,000 (88.0%) and 300/300 (100%) on evaluation sets used during development (VENDOR-REPORTED). [source](<https://huggingface.co/togethercomputer/Tev1-4B-experimental>)
- **INDEPENDENT**: In jev-rsi's independent baseline test, the author measured untuned Qwen3.5-4B using [Tev1](<https://decisionmodels.ai/models/tev1>)'s decision protocol through local Ollama at 673/1,000 (67.3%) on test and 130/300 (43.3%) on policy\_transfer. [source](<https://github.com/Joshhuang123/jev-rsi>)

## Links

- [Source: Hugging Face](<https://huggingface.co/togethercomputer/Tev1-4B-experimental>)
- [Source: GitHub](<https://github.com/togethercomputer/tev1>)
- [Source: Announcement](<https://x.com/togethercompute/status/2102882216950763814>)
- [Source: Untuned 4B baseline test](<https://github.com/Joshhuang123/jev-rsi>)

## Related projects

### Local intent router with a model cascade

[Local intent router with a model cascade](<https://decisionmodels.ai/projects#project-hybrid-intent-router>)

[Source: indranildchandra/hybrid-intent-router](<https://github.com/indranildchandra/hybrid-intent-router>)

Routes support messages through regex, then a small classifier, then decision models, then an LLM as a last resort, all running locally. [Laya](<https://decisionmodels.ai/models/laya>) answers the guardrail and department questions, and [Tev1](<https://decisionmodels.ai/models/tev1>) is the default for the next tier.

- Creator: Indranil Chandra

- Stars: 0

- Licence: MIT

- Model: [Laya](<https://decisionmodels.ai/models/laya>), [Tev1](<https://decisionmodels.ai/models/tev1>)

- Tags: routing, local

### Pacman played by Tev1 in Ollama

[Pacman played by Tev1 in Ollama](<https://decisionmodels.ai/projects#project-pacman-systemone>)

[Source: bidouilles/pacman-systemone](<https://github.com/bidouilles/pacman-systemone>)

A one-file browser Pacman where the open directions at each step become the options in a single question to a local [Tev1](<https://decisionmodels.ai/models/tev1>) model through Ollama's decision endpoint.

- Creator: bidouilles

- Stars: 0

- Licence: no licence

- Model: [Tev1](<https://decisionmodels.ai/models/tev1>)

- Tags: game, ollama

### Thunderbird email tagging that learns

[Thunderbird email tagging that learns](<https://decisionmodels.ai/projects#project-kestrel>)

[Source: sarath/krestel](<https://github.com/sarath/krestel>)

A Thunderbird extension that tags your email with decision models served by Ollama's /v1/systemone endpoint, such as [Tev1](<https://decisionmodels.ai/models/tev1>) 0.8B, [Tev1](<https://decisionmodels.ai/models/tev1>) 4B or [Nimble](<https://decisionmodels.ai/models/nimble>), and keeps learning from the tags you add by hand.

- Creator: sarath

- Stars: 0

- Licence: MIT

- Model: [Tev1](<https://decisionmodels.ai/models/tev1>), [Nimble](<https://decisionmodels.ai/models/nimble>)

- Tags: email, local