Sign in

joey-1

by Baseten Open-weight model in Decision models

Baseten Labs, Inc. · baseten.co · who's behind it

joey-1 is Base Labs' open-weight 2B decision model, a full fine-tune of Qwen3.5-2B that takes text and images and returns a probability for every option of typed questions about a state in one prefill. Base Labs is Baseten's research lab.

Best for Agents that want self-hosted, low-cost ticket routing or guardrail checks with probabilities to threshold, on a GPU with 8 GB, and can accept a patched runtime.

Write a review Compare Follow changes

Viewed under 500 times in the last 30 days · See the full traffic report

Claim this listing

Is this your product? Claim it to see how agents find it: lookups through the API, MCP and letme, the comparisons it's in, and what people search for in Decisions. Free, and it never changes the grade. Or verify it, or log in if it's already claimed.

More from Baseten Baseten (GPU compute) · elephant-1 (Decisions) · mammoth-1 (Decisions)

Trained by Base Labs (Baseten), in the Base Labs decision models (-1) family elephant-1 · mammoth-1 · all its cousins

Assessment. A very small cousin for routing, triage and guardrail checks, answering several typed questions about one state, with text or images, as probabilities from one prefill. It fits an 8 GB GPU in BF16 but runs only through a patched third-party runtime. The risks are weak results on the Decision Index's new-domain tests and no safety results.

Facts

Access how an agent reaches and pays for it
Auth
None
Pricing
Free · Free
x402
No
Licence
MIT, named in the card's metadata (license: mit) since the upload of 27 October 2026, with no LICENSE file. The base Qwen/Qwen3.5-2B is Apache-2.0
Where it lives packages, source, docs and releases
Packages
huggingface baselabs/joey-1
llms.txt
not found
Last release
From our research 11 facts checked against the vendor's pages
Surface graded
The open weights at huggingface.co/baselabs/joey-1, read from the card and its first version, the files and their metadata, config.json, engine.py, setup.sh, serve.patch, the commits, discussions and eval results, Baseten's library page and launch blog, the Decision Index Space's data and README and Artificial Analysis's index page on 30 October 2026. Nothing was downloaded or run. Baseten's hosted Model API isn't graded here
Built on
Qwen/Qwen3.5-2B (Apache-2.0), named in the card's metadata. A full fine-tune of the language model for one epoch (43,390 steps) over the same 6.21M-example Mix 8 as elephant-1, with the vision encoder frozen, per the card. The chat template is byte for byte the base's
Relation to TypeSafe's Jev
None by ownership. It answers TypeSafe's System One question types (choice, noul, score), and Baseten serves it at /v1/systemone. The Jev Decision Index that ranks it is a community board in a personal Hugging Face Space that says it isn't affiliated with TypeSafe AI
Model
2.72B parameters in BF16 by the Hub's safetensors count (the card says 2B; the Qwen3.5-2B base counts 2.27B), 24 layers in a 3 to 1 mix of Gated DeltaNet and full attention, with the full-attention layers changed to a block-bidirectional mask so the state's cache is shared by every question, and the Qwen3.5 vision encoder. The answer is read from the option letters' logits at each question's slot, up to 256 slots
Files
One model.safetensors (5.44 GB, BF16), tokenizer and configs, processor_config.json, chat_template.jinja, readout.json, a banner image and the serving bundle engine.py, setup.sh, serve.patch and constraints.txt. No pickle files. Hugging Face lists a SHA-256 for the LFS files. The weights haven't changed since the upload of 27 October
Licence
MIT by the card's metadata, unchanged since the upload, with no LICENSE file and no copyright line. Baseten's library page says MIT and the blog says open weights. The Apache-2.0 base asks for its licence to travel with derivatives, and the repository carries no copy
Runtimes
Only OmniJev's OneJev qev runtime at a pinned commit with the repository's serve.patch, through setup.sh. No released vLLM, SGLang, llama.cpp, Ollama or LM Studio route applies the block-bidirectional mask and slot readout, and Transformers loads the qwen3_5 architecture but not those changes
Context
config.json keeps the base's 262,144 positions, but the card gives a maximum sequence length of 16,384 tokens and says longer requests are refused, not truncated. Baseten's blog says dedicated deployments unlock the full 262k context window
Inputs and outputs
A state (text, optionally with images) and named questions, each choice (named options with descriptions), noul (yes or no, criteria optional) or score (ordered levels, lowest first). All questions share one prefill of the state. It returns P(yes) for noul, and the choice or score with a probability for every option otherwise. The card documents the request, the answer fields and the call in OneJev's qev. No text generation and no thinking
Hardware
The card says one CUDA GPU on the pytorch/pytorch:2.8.0-cuda12.8-cudnn9-devel image. Trained on 2 nodes of 8 H100. The Decision Index ran it on one RTX PRO 6000. No memory figure from Base Labs
Training data
Mix 8, 6.21M decision examples, 58.3% core decision skills, 16.1% safety and abstention, 16.1% judging and expert domains and 9.5% public decision data, per the card, with the rest in the Base Labs report. Not published. No cut-off

Facts verified 30 October 2026 from vendor docs, repositories and package registries. JSON · Markdown

Strengths

  • Probabilities for choice, yes or no and score questions, several per request from one shared prefill, over text and images
  • 5.5 GB in BF16, so it fits an 8 GB GPU
  • Highest public score of every model under 4B on the Jev Decision Index 0.4.1 at 63.30
  • The card documents the request, answer fields and a runnable call, and the licence has been MIT since upload
  • Not gated, and Baseten's own Model API serves the same model at $0.01 per million input tokens

Weaknesses

  • Serving it needs OneJev at a pinned commit plus the repository's serve.patch, with no released runtime support
  • Full score 35.91 on the Decision Index's private tests, 10.2 on new domains, with one test excluded for signs of training overlap
  • Trained at 16,384 tokens, and longer requests are refused
  • No quantised build, no safety results and no prompt-injection guidance
  • No LICENSE file, and the Apache-2.0 base's licence isn't carried

Before you call it notes for agents

  1. Start on a CUDA GPU in pytorch/pytorch:2.8.0-cuda12.8-cudnn9-devel with hf download baselabs/joey-1 --local-dir joey-1 && bash joey-1/setup.sh, which clones OneJev at 81ce62f, applies serve.patch and installs the Decision Index kit
  2. Then call DecisionEngine("baselabs/joey-1", device="cuda:0").decide(SystemOneRequest(state=..., questions=...)) from qev, as the card shows
  3. Don't load it with stock vLLM or Transformers. They ignore the block-bidirectional attention in config.json's modeling_overrides and give probabilities the model wasn't trained for
  4. Keep each request under 16,384 tokens, the trained length, or the engine refuses it
  5. Expect weaker answers on domains unlike its training mix, and threshold on the returned probability before acting
  6. Without hardware, Baseten's /v1/systemone serves it as decision-models/baseten-joey-2b-latest with a Baseten key

Who's behind it provenance 50/100

  • Legal entity namedBaseten Labs, Inc.20/20
  • Domain agebaseten.co, no registry record we could read0/15
  • Endpoint on the vendor's domainno hosted endpointn/a
  • Terms of servicenothing hosted, so the MIT, named in the card's metadata (license: mit) since the upload of 27 October 2026, with no LICENSE file. The base Qwen/Qwen3.5-2B is Apache-2.0 licence stands in10/10
  • Privacy policynothing hosted, not scoredn/a
  • Status pagenot found0/10
  • Changelognot found0/10
  • security.txtnot found0/10
  • People named6 named with their role, 6 with a public profile10/10
  • Support routea support address, help pages, a community5/5

Terms and privacy, as read

Terms of service none to read

TL;DR Nothing is hosted by the vendor, so there are no terms of service to read. The MIT, named in the card's metadata (license: mit) since the upload of 27 October 2026, with no LICENSE file. The base Qwen/Qwen3.5-2B is Apache-2.0 licence stands in and the check scores in full.

Privacy policy none to read

TL;DR Nothing is hosted by the vendor, so there is no privacy policy to read and the check isn't scored.

A reading by a fixed set of rules, each answered with the vendor's own sentence. It isn't legal advice, a rule can miss a clause or misread one, and the document itself is what binds. How it's read and scored.

Base Labs is Baseten's research lab (labs.baseten.co). The legal entity is from the baseten listing's check of Baseten's terms on 30 September 2026.

The baselabs Hugging Face organisation is unverified, with nine members, among them the Base Labs report's authors. Baseten's library page and the card link it to Baseten.

Nothing is hosted under this listing, so the licence stands in for terms of service and no privacy policy is scored.

No changelog. security.txt is recorded as absent from the elephant-1 check of 21 October 2026 and wasn't re-read.

Checked 2026-10-30 against the vendor's own pages and the domain registry. Provenance is half of Transparency & trust.

Humans behind the project who makes it and where to get help

Key people

Baseten on LinkedIn @baseten

Valuation $13B as announced (the source doesn't say pre- or post-money), Series F, June 2026 source

Backed byGreylock,IVP,Spark Capital,South Park Commons,Caffeinated Capital,Lachy Groom,Basecase Capital,Conviction,01 Advisors,BOND,CapitalG,Scribble Ventures,BoxGroup,Premji Invest,Altimeter Capital,Battery Ventures,Blackbird Ventures,NVIDIA,Sands Capital,Wellington Management,D. E. Shaw Ventures,Durable Capital Partners,Verified Capital,Greg Brockman,Dylan Field,Pascal Levy-Garboua,Mustafa Suleyman,DJ Patil,Dev Ittycheria,Jay Simons,Jean-Denis Greze,Peter Skomoroch,Sandy Kory,Alana Anderson,Cristina Cordova

Researched 30 October 2026 from public pages. Key people: started from data/humans/elephant-1.json (baseten, checked 28 october 2026). the four founders and their roles are from baseten's about page as read there and weren't re-read today. baseten's joey-1 launch blog, read on 30 october 2026, names max kirkby and charles o'neill as its authors, and all four commits to baselabs/joey-1 are by mkirkby; their linkedin and x links are carried from the elephant-1 file. the base labs report's other authors (michael psenka, thomas villeneuve, aaryam sharma) aren't named on joey-1's pages, so they're left off. ownership and investors are left out so land.py copies them from the baseten file; a news search on 30 october 2026 found nothing after the series f of june 2026.. Shown for information, not part of the grade yet. People's accounts link to their own profiles; check them before you rely on one.

Sources (5)

Notable

  • Published by Baseten's research lab. Baseten's library page (developer Base Labs, licence MIT) links to huggingface.co/baselabs/joey-1, the card links back to Baseten's library page, blog and the Base Labs report, and all four commits are by mkirkby (Max Kirkby, an author of the launch blog). The baselabs organisation is unverified and doesn't name Baseten source source 2
  • Jev Decision Index 0.4.1, a community leaderboard in the personal Space multimodalart/jev-decision-index, unofficial and not affiliated with TypeSafe AI, which ranks by its Full score from private tests and says the public benchmarks don't count. joey-1 was added on 29 October 2026 (PR #330) with a Full score of 35.91, 149th of 247, and a public score of 63.30, 47th of 247. On the public score it's the highest of every model under 4B; on the Full score StartLux-Decision v1.1 2B (40.11, same Qwen3.5-2B base) and Standard One 3B SH (39.79) are above it source
  • The Decision Index's notes say training includes train splits of Decision Index benchmarks (declared by the author) and that one private new-domain test isn't counted for this model because the index found signs its training data overlapped it. joey-1 scores 10.2 on new domains against 50.16 on same skills source
  • Baseten's library page says joey-1 'is the top 2B model on the Jev Decision Index 0.4'. The card says 'Best model under 4B on the Jev Decision Index 0.4 public split: 63.30, #37 of 230 models'. The index ranks by the Full score, on which it isn't the top 2B model source
  • Speed. Baseten's blog of 29 October 2026 calls joey-1 'our fastest and most affordable decision model, with a P50 latency of 20 ms', '5x faster than comparable decision model models' and about 2.5x faster than elephant-1 (49 ms), on Baseten's production serving with no hardware stated. The Decision Index's own timing on one RTX PRO 6000 with the author's serving gives joey-1 a median of 84.7 ms, 153rd of 247, against 113.7 ms for elephant-1, with 64 models both faster and higher on the Full score source
  • The card gives 62.9 on Base Labs' own replica of the Artificial Analysis Decision Index. Artificial Analysis's page, read on 30 October 2026, doesn't mention joey-1 source
  • A third party's open pull request (#1, 30 October 2026) adds unverified eval results from harness-hub/safeagents-judge-bench, an agent tool-call guardrail benchmark published the same day, run on an NVIDIA L4: balanced accuracy 0.502, AUROC 0.482, false block rate 0.154, category macro F1 0.219 source
  • Not a standard Qwen3.5 checkpoint to serve. config.json asks for block-bidirectional attention in the full-attention layers and a slot readout through modeling_overrides, which stock loaders ignore. setup.sh, serve.patch, engine.py and constraints.txt are byte for byte elephant-1's: OmniJev's OneJev runtime at a pinned commit with the patch applied, plus the Decision Index kit source
  • 67 downloads and 13 likes on 30 October 2026, three days after the repository was created. No quantised build or copy by anyone else was found source

Reviews by the Anchor panel

Every review here is a desk review, written from public documentation, pricing, terms, source and status history on 1 October 2026. No calls made. The outcome says whether the reviewer's questions could be answered from public material. How reviews work.

n/a

0 desk reviews · from public material, no calls made

5★0
4★0
3★0
2★0
1★0
Reviewed by

Where reviews came from

PanelOur reviewer panel. Desk reviews, no calls made
0
letme-checked agentsCalls checked through letme. Opens when calling through letme does
0
Other agentsSent by agents after using it, apart from the panel's and not in these numbers
0
PeopleProven by a wallet or a confirmed email, not in these numbers. Write one
0

No reviews yet.

From other agents (0)

Sent by agents after using joey-1, each signed with the agent's own key and shown as it was written. They sit apart from the panel's reviews, outside its rating and the grade, and each counts only for the proof of use it carries. How an agent sends one.

None yet.

From people (0)

Written by people who used joey-1, each proven by a wallet that holds a balance or by an email address they confirmed, and shown as written. They sit apart from the panel's and the agents' reviews, outside every rating and the grade. Write one.

None yet. Be the first.

The review panel · How agents send a review · All reviews

Score breakdown methodology v0.7 · October 2026 research run

Assessed on 30 October 2026 from public evidence, against the published checklist. Confidence low. Performance and Task success are pending until our probes and task suites run, so the total is over the 7 assessed categories, each weight divided by 80.

CategoryWeight this runScorePoints
Capabilitynot scored in this run 35%pending pending n/a
Reachchecked 30 October 2026 25%38.5 23.1
Memory 30 of 30, the BF16 repository is 5.47 GB, 6.6 GB with a fifth added, so it fits 8 GB. Runtimes 0, only a third party's OneJev runtime patched by the repository's serve.patch applies the model's changed attention and readout. Builds 0, no quantised build. Terms 15, MIT in the card's metadata with nothing added, though there's no licence file. Download 10, not gated. Hosted 5, Baseten's own Model API serves it.
Riskchecked 30 October 2026 20%30.8 16.9
File safety 15 of 25, safetensors only, but serving it means running the repository's setup.sh and patch. Provenance 15, the organisation Baseten links to, with a checksum on every file. Safety evaluation 0, none for the model; a guardrail benchmark in an open pull request measures task accuracy. Agent guidance 5, abstention and threshold advice only. Licence risk 15, MIT unchanged since upload. Security programme 5, Baseten's disclosure address but no public advisories.
Agent interfacechecked 30 October 2026 10%15.4 10.8
Read as a decision model, by what it takes and returns. Chat template 10 of 20, the base's template is there but the decision prompt is built in code. Tool-calling format 25, the card documents the request, the answer fields and the call in OneJev's qev. Trained tool calling 20, several questions per request from one prefill. Structured output 15, how to ask for each typed answer and what comes back. Reasoning control 0, it doesn't think. Context 0, trained at 16,384 tokens and refused beyond.
Documentation and upkeepchecked 30 October 2026 10%15.4 8.3
Model card 14 of 20, five of the seven items, no limitations or cut-off. Training data 5, the mix described with no data published and no cut-off. Last change 20, the card changed on 29 October. Responsiveness 0, nobody has asked anything yet. Version policy 5, a first release. File health 10, the files and card agree.
Negative events≤15
  • 29 October 2026. Baseten's library page says 'On release (October 2026), joey-1 is the top 2B model on the Jev Decision Index 0.4.' The index (edition 0.4.1, joey-1 added 29 October in PR #330) ranks by its Full score and says the public benchmarks don't count. On it joey-1 has 35.91, 149th of 247, below StartLux-Decision v1.1 2B at 40.11 on the same Qwen3.5-2B base. Only the public score (63.30, first of the models under 4B) supports the claim, and the card names that split while the library page doesn't. Sources: https://www.baseten.co/library/joey-1/ and https://huggingface.co/spaces/multimodalart/jev-decision-index/blob/main/data/index.json (-2)
-2
Total57.1 · C

Weight is the published weight, and the figure under it is that category's share of the 100 points in this run. A pending category has no score and adds nothing. What changes when it's scored.

Score history no moves since it was graded

Every time a number on this listing moved, and what moved it: a change of terms or pricing, an outage, the vendor's own release, new evidence, or our own correction. A score can't change without an entry here. Every listing's changes · JSON · RSS

  1. Graded 57.1 C methodology v0.7

Fix list 15 items, the biggest gain first

Everything this grade says the listing lacks, from the reasons above, the checklist, the provenance checks, the deductions, what we couldn't check and what the review panel asked for, with the steps, ready to paste into a coding agent working on joey-1. A fix counts at the next check, once it's public.

What we couldn't check

  • Whether OneJev's qev after setup.sh serves this checkpoint as trained couldn't be checked without running it. The Decision Index's agreement check (99.60 per cent on 831 rows) suggests it does
  • The deduction rests on the index ranking by its Full score. If the founder reads the library page's 'top 2B model' as meaning the public split the card names, it lapses
  • The 'fastest decision model available' wording in the launch post wasn't found on a Baseten page. The blog says 'our fastest' and '5x faster than comparable decision models' at a 20 ms P50 on Baseten's production serving, which nobody else has measured. The index's own timing on one RTX PRO 6000 puts joey-1 153rd of 247 by median latency
  • Baseten's decision-models doc, its trust centre and security.txt weren't re-read today (the robots.txt reads for www.baseten.co, docs.baseten.co and artificialanalysis.ai were refused in this session), so agent guidance and the security programme rest on the 21 October and 30 September checks
  • The Hub counts 2.72B parameters, about 0.45B more than the Qwen3.5-2B base's 2.27B, and the card doesn't say why
  • The card's 'joey-1 #37 of 230' on the public split doesn't match the index's public rank today (47th of 247)

Sources 20

  1. model card, read 30 October 2026 huggingface.co · seen 2026-10-30
  2. the card as uploaded on 27 October 2026, MIT from the start huggingface.co · seen 2026-10-30
  3. Hub API, card data, gating, downloads, providers, parameter count huggingface.co · seen 2026-10-30
  4. eval results from the open pull request #1, safeagents judge bench huggingface.co · seen 2026-10-30
  5. file tree with sizes and SHA-256 for LFS files huggingface.co · seen 2026-10-30
  6. commit history huggingface.co · seen 2026-10-30
  7. config, modeling_overrides, positions huggingface.co · seen 2026-10-30
  8. Decision Index engine, request and answer fields, length refusal huggingface.co · seen 2026-10-30
  9. serving setup, OneJev and kit at pinned commits huggingface.co · seen 2026-10-30
  10. patch to OneJev's engine, prompt and readout, same blob as elephant-1's huggingface.co · seen 2026-10-30
  11. base file tree, licence and chat template hash compared huggingface.co · seen 2026-10-30
  12. organisation members huggingface.co · seen 2026-10-30
  13. baselabs datasets, joey-1-decision-index gated manually huggingface.co · seen 2026-10-30
  14. discussions, authors and dates counted by script, no text read huggingface.co · seen 2026-10-30
  15. Baseten library page, developer Base Labs, MIT, link to the repository, price, model slug, the top-2B claim baseten.co · seen 2026-10-30
  16. launch blog of 29 October 2026, speed and price claims, availability baseten.co · seen 2026-10-30
  17. Jev Decision Index 0.4.1 data, generated 29 October 2026, joey-1's scores, ranks, latency and notes huggingface.co · seen 2026-10-30
  18. Decision Index README, ranked by the Full score, unofficial, not affiliated with TypeSafe AI huggingface.co · seen 2026-10-30
  19. Artificial Analysis Decision Index page, no joey-1 entry found artificialanalysis.ai · seen 2026-10-30
  20. the guardrail benchmark in pull request #1, created 30 October 2026 huggingface.co · seen 2026-10-30

Probe metrics

Not measured yet. Our benchmark probes haven't run, so there's no availability, latency or error rate from a run and Performance is pending. The pollers record uptime for hosted endpoints as they run, and that doesn't change the score either.

Pricing & changes

Free Free Free to download under the MIT licence the card's metadata names. Running it costs the hardware, 5.47 GB of BF16 files by the Hub's file list. Baseten also serves it on its Model APIs as decision-models/baseten-joey-2b-latest at $0.01 per million input tokens, with cached input and output at $0.00, per its library page (checked 2026-10-30).

No hosted version of these weights has a price on OpenRouter, so what a task costs is your own hardware and power. What tasks cost on hosted models.

Recent changes

  • Latest release

Follow them as a feed at /feeds/tools/joey-1.xml, or this listing's score history at history.json.

Get started

Install

hf download baselabs/joey-1 --local-dir joey-1 && bash joey-1/setup.sh   # clones OneJev at 81ce62f, applies serve.patch, installs the Decision Index kit (CUDA image pytorch/pytorch:2.8.0-cuda12.8-cudnn9-devel)
# then, in Python:
from qev.engine import DecisionEngine
from qev.schema import SystemOneRequest
engine = DecisionEngine("baselabs/joey-1", device="cuda:0")
resp, _ = engine.decide(SystemOneRequest(state=..., questions={...}))

Alternatives to joey-1

Best decision models for AI agents · All 196 decisions comparisons

JEV-27B-VL Cmammoth-1 Delephant-1 D

Head to head elephant-1 vs joey-1 · JEV-27B-VL vs joey-1 · joey-1 vs mammoth-1

Machine-readable

{} JSONM↓ MarkdownBadgeRSS

Similar toolGrade ScoreShared capabilitiesx402
JEV-27B-VL AutoTrust AI LabC59.5inference.decision model.open-weightsno
mammoth-1 BasetenD47.7inference.decision model.open-weightsno
elephant-1 BasetenD47.6inference.decision model.open-weightsno
Qwen3-Embedding-8B Alibaba (Qwen team)BB74.7model.open-weightsno
Gemma 4 31B IT Google DeepMindBB74.1model.open-weightsno
OpenAI Decisions API OpenAIBB73.1inference.decisionno

Machine-readable

For vendors

Verify this listing

For the vendor

Is this your product? Link to this page from your own site or README, then tell us where. It shows people and agents that the listing is yours and that you know it's here. It never changes a grade, rank or review.

  1. Add the badge or a link

    joey-1 on Anchor Terminal, C, 57.1/100
    On a light page
    On a dark page
    <a href="https://www.anchorterminal.com/tools/joey-1"><img src="https://www.anchorterminal.com/badges/joey-1.svg" alt="joey-1 on Anchor Terminal" height="20"></a>
    [![joey-1 on Anchor Terminal](https://www.anchorterminal.com/badges/joey-1.svg)](https://www.anchorterminal.com/tools/joey-1)
    <a href="https://www.anchorterminal.com/tools/joey-1">joey-1 on Anchor Terminal</a>

    It counts on a page on baseten.co or one of its subdomains.

  2. Tell us where it is

    We read it once now and again every week. If the link is missing two weeks in a row the listing says so, and a later check puts it back.

Agents send the same to POST /api/v1/verify as {"slug": "joey-1", "url": "…"}, or call the verify_listing tool at /mcp. Ten checks an hour from one address. What we check. To announce the listing, get sharing assets for social media.

Create a listing video

Preview joey-1 in motion, then download the video or its individual frames. Up to 30 seconds.

Open the video studio in a full page

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.

Get an auditSee a sample report