Decision ModelsDecision Models
Get API key

Decision Models · Jev-compatible API

Route, approve, score.
One fast API call.

Send a state and a question, get one answer per option, with probabilities. Open-weights decision models behind a Jev-compatible API: s1-fast, s1-pro and s1-vision.

Get an API key Read the quickstart

Sign in with Google or a one-time email link. Top up from EUR 1; taxes are shown before you pay. Operated by a German company in Passau: imprint.

  • 85.2 vs 75.0
    JevBench capability score: mammoth-1 (s1-pro in the EU tier) vs Jev 1.14.0, same items1
  • $0.019 €0.017s1-fast, per 1M input tokens. Output tokens are free.
  • EUPersonal data? The EU tier keeps requests in the EU, with a DPA at signup.
Request → responseIllustrative samples
REQUEST / choice

state: Mia owns a red bicycle.

question: Which color is the bicycle?

ANSWER / probabilities
red0.91
blue0.09
REQUEST / choice

state: a business customer asks to update an invoice address.

question: Which support queue fits?

ANSWER / probabilities
billing0.87
technical0.09
account0.04
REQUEST / noul

state: a viewer deletes a workspace.

question: Is this action allowed?

ANSWER / P(true)
true0.04
Illustrative contract shapes, not measured output — all samples and copyable code are in the quickstart.

Measured by us on 11–12 Oct 2026; method and raw data on /evidence.

1 mammoth-1 (s1-pro in the EU tier) scores 10.2 points above Jev 1.14.0 on the same items. JevBench v1.8.7 (4 Nov 2026) capability score = mean of Intelligence and Calibration; each pair was measured on the same items. JevBench measures the open models, not our API. JevBench is run by Benchmark Heaven, also by the founder of Decision Models; method and raw results are public. Evidence

Quickstart

Your first decision in three steps.

Jev-compatible request shape: moving from Jev means changing the base URL and credentials, then reviewing the documented differences.

  1. Get a key

    Sign in with a one-time email link or with Google and create an API key in the dashboard. Free allowance by invitation; top up from EUR 1.

  2. Send a question

    A named state and up to three questions for s1-pro, one for s1-fast or s1-vision: choice, score or noul (a yes/no question answered with a probability).

  3. Use the answer

    Read the chosen answer with probabilities, a usage receipt and the effective tier. Choose the tier per API key or per request with the S1-Region header.

Request · POST /v1/systemone
# Key from your dashboard
export DECISIONMODELS_API_KEY="paste-your-key-here"
# Ask one question on the Peer-to-Peer tier (S1-Region: global)
curl https://api.decisionmodels.io/v1/systemone \
  -H "Authorization: Bearer $DECISIONMODELS_API_KEY" \
  -H "Content-Type: application/json" \
  -H "S1-Region: global" \
  -d '{
  "model": "s1-fast",
  "state": "Mia owns a red bicycle.",
  "questions": {
    "color": {
      "type": "choice",
      "instructions": "Which color is the bicycle?",
      "criteria": {"red": null, "blue": null}
    }
  }
}'
Get an API key Try the playgroundFull quickstart →

Two tiers, one API

Choose where your decisions run.

GLOBAL · PEER-TO-PEER

Compute from everyone, for everyone.

Our lowest price per decision, on capacity that grows with demand. Peer-to-Peer capacity depends on how many network providers are live, so short bursts can get a retryable 503; for steady high volume use an EU key.

Today: spare EU capacity first, then rented worldwide GPU capacity from independent hardware providers for keys that opt in to worldwide processing.

Network processing is preselected for new Peer-to-Peer keys ("Our GPUs only" is selectable); independent providers see request content in clear text and may store it. Send no personal data on these keys.

Switchable per keyHow peer-to-peer works →

EU TIER

Hosting in EU.

Built for GDPR — DPA accepted at signup; our gateway does not log prompts or answers, and we do not use them for training.

EU AI Act statement — model cards, benchmarks and intended use documented.

EU requests always stay in the EU and get priority.

EU tier details →

EU tier

For personal data: the EU tier.

Same models, same API. One header or an EU key is all it takes.

Explore the EU tier Read the DPA →

What you can do with it

Small decisions across your workflows.

Send context and a named question; use the typed answer in your workflow.

Support

Route tickets to the right queue

Map each request to the team that can handle it.

Invoice address update → billing | technical | account

Support

Spot cases that need escalation

Flag sensitive requests using your escalation policy.

Charge dispute → escalate?

Product / Safety

Review actions against policy

Check a proposed action. Keep deterministic access controls in your application.

Viewer deletes workspace → allowed?

Quality

Score drafts against a rubric

Grade replies or summaries and escalate uncertain cases.

Draft + rubric → score

Sales

Prioritise inbound leads

Sort messages by buying intent using your criteria.

Rollout request → hot | warm | cold

Ops / Docs

Send documents to the right workflow

Classify document text before handing it off.

Document → invoice | contract | other

Agents

Choose the next useful step

Route between search, clarification, response and stop.

Context + tools → next step?

LLM-cost routing

Choose where to spend LLM compute

Classify difficulty and stakes before selecting a model.

Request → category · difficulty · stakes

Illustrative workflows. Validate representative cases on your own data before automating decisions.

Beyond our API

Use any open decision model.

Pick a model from the JevBench open-weights leaderboard and run it the way you need.

Dedicated hosting

Your own private endpoint for one model, EU or global.

Learn more →

Run locally

Installer that checks your hardware and downloads the model. Free for individuals and small teams.

Learn more →

Pre-installed hardware

A ready-to-run box for your office.

Learn more →

Browse all models →

Model lineup

REAX 1.3

reax-1.3 · REAX 1.3 by Decision Models · text · EU only

REAX is an open inference network on Bittensor subnet 92; JevBench is a sister project. Its open training competition improves the decision models served by the network. REAX 1.3 has two parts that run in parallel on one GPU; each part builds on a fine-tune trained by a REAX miner (uid 18 and uid 83). Jev-compatible interface, with a probability for every option. Text only, 1 question per request, up to 28,672 billed input tokens including the state; longer inputs are refused, not truncated. Hosted in the EU tier only. €0.400 per 1M input tokens, output free.

Open weights on Hugging Face (Apache-2.0) · REAX Engine (MIT) · Model page →

s1-pro

mammoth-1 27B by baselabs · MIT · FP8 · text · EU tier

Since 5 November 2026, mammoth-1 (revision 61cec44c) answers s1-pro in the EU tier. Up to three text questions per call, 8,192 input tokens per question. In the EU tier, s1-pro is answered by mammoth-1 (s1-pro-5), or by s1-pro-3 (vLLM-SR d3) if mammoth-1 cannot answer. In the Peer-to-Peer tier, it is answered by s1-pro-3, with or without network processing enabled. The pinned id s1-pro-5 is EU tier only. model_version shows which version answered. mammoth-1 is post-trained from Qwen3.5-27B by the Qwen team (Apache-2.0). Not affiliated with or endorsed by baselabs, Baseten or the Qwen team.

s1-pro-3

vLLM-SR d3 · 26B · FP8 · Apache-2.0

The pinned id for vLLM-SR d3 26B (Apache-2.0), served in FP8. It has answered s1-pro since 12 October 2026; since 5 November 2026 it is the Peer-to-Peer model and the EU fallback. Up to three questions per call, 16,384 input tokens per question. Available through at least 27 January 2027. The authors do not disclose its training data.

Hugging Face: vllm-sr/d3

s1-pro-5

mammoth-1 27B · MIT · FP8 · pinned id · EU tier only

The pinned id for mammoth-1 (revision 61cec44c), EU tier only. Served in FP8 by Decision Models, up to three questions per call, 8,192 input tokens per question. Billed at the s1-pro EU list rate: $0.084 per 1M input tokens, output tokens free. mammoth-1 is an open model under MIT, post-trained from Qwen3.5-27B by the Qwen team (Apache-2.0). Not affiliated with or endorsed by baselabs, Baseten or the Qwen team.

Hugging Face: baselabs/mammoth-1

s1-pro-2

Perplexity Decider v1.1 27B · NVFP4 · previous s1-pro, pinned id, retires 1 December 2026

The pinned id for the model that answered s1-pro until 12 October 2026: calibrated text decisions on our engine, at about 4.4x the throughput of the author's reference code (not Perplexity's hosted service) on the same GPU model (measured by us on 11 October 2026; method on /evidence). Available in both tiers, billed at the s1-pro list rate: $0.080 Peer-to-Peer / $0.084 EU per 1M input tokens, output tokens free. Perplexity Decider v1.1 is an open model under Apache-2.0, derived from Qwen3.8-27B by Alibaba Cloud (Apache-2.0). Retires on 1 December 2026; successor: s1-pro.

Hugging Face: perplexity-ai/pplx-decider-v1.1-27b

s1-fast

Based on Blink v0.3 26B-A4B · NVFP4 · text

Short text decisions at low latency: one question with up to 16 options. Runs on a batched engine, so probabilities for identical requests can differ slightly.

s1-vision

Based on Blink v0.3 26B-A4B · NVFP4 · one image

One question with one image attached, up to 26 options.

reax-1

REAX 1 by Decision Models · 27B · text

Our own open decision model: REAX 1 by Decision Models, a fine-tune of Perplexity Decider v1.2 by Perplexity (Apache-2.0), derived from Qwen3.8-27B by the Qwen team (Apache-2.0). Not affiliated with or endorsed by Perplexity AI or the Qwen team. Up to three questions per call, billed at the s1-pro list rate. reax-1 always points to the current REAX 1 release. We may replace the model behind this id with a newer REAX version at any time, with the same API and price; we switch only to a version that measures at least as well on our evaluations. For reproducibility, pin a dated id such as reax-1-20261021; pinned ids stay available for at least 30 days after they stop being current.

5 November 2026: s1-pro now runs mammoth-1 (s1-pro-5) in the EU tier. In the EU tier, s1-pro is answered by mammoth-1 (s1-pro-5), or by s1-pro-3 (vLLM-SR d3) if mammoth-1 cannot answer. In the Peer-to-Peer tier, it is answered by s1-pro-3, with or without network processing enabled. The pinned id s1-pro-5 is EU tier only. model_version shows which version answered. Prices unchanged.

Every response names the model version that answered in model_version and in the S1-Model-Version header. The retired ids s1-pro-1, s1-fast-1 and s1-vision-1 are answered by the current models. In the EU tier, s1-pro is answered by mammoth-1 (s1-pro-5), or by s1-pro-3 (vLLM-SR d3) if mammoth-1 cannot answer. In the Peer-to-Peer tier, it is answered by s1-pro-3, with or without network processing enabled. The pinned id s1-pro-5 is EU tier only. model_version shows which version answered. Perplexity Decider v1.1 stays pinned as s1-pro-2 through 30 November 2026. s1-pro-2 and s1-pro-4 retire on 1 December 2026; s1-pro-3 stays available through at least 27 January 2027. On keys with peer-to-peer network processing enabled, s1-fast-1 may instead be answered by its original model (listed in the model licences). Each question, including its state, is limited to 8,192 to 28,672 input tokens, depending on the model. Full model list: /models · Licence.

Capability, price and speed

Current prices are on /pricing. Dated Decision Index results and our own speed measurements for s1-pro-3, s1-pro-2, s1-fast and s1-vision are on /evidence; reference-model results from published benchmarks and their methods are on /benchmarks.

Pricing

Pay per input token. Output is free.

Free allowance by invitation. Top up from EUR 1.

ModelPeer-to-PeerEU
s1-fast$0.019
€0.017		$0.021
€0.018
s1-pro$0.080
€0.070		$0.084
€0.074
s1-vision
text without an image		$0.031
€0.027		$0.032
€0.028

Per 1M input tokens, prices for new customers, excluding applicable taxes. Output tokens are free. All prices, image rates and calculator →

Where it fits

A fast decision API for narrow tasks.

Use a decision model for narrow tasks where it meets your accuracy needs. Keep an LLM for difficult reasoning and open-ended writing.

TYPICAL TASKS

Built for narrow decisions

Route a ticket, approve an action or score a draft: one answer per option, with probabilities.

WHEN AN LLM IS BETTER

Know the limits

Use a frontier LLM for open-ended writing or difficult reasoning. Validate representative cases on your own data first.

Press context

What TechCrunch wrote about Jev

“…LLMs as we know them aren't the right solution for a lot of software because they are comparatively slow and expensive.”
— TechCrunch; the article is about Jev, not Decision Models.

Under the hood

The REAX Engine

Our prefill-only inference engine for decision models (one forward pass + a readout over the answer options). On one RTX PRO 6000 (FP8 mode) it serves Perplexity Decider v1.1 at about 3.1x the decisions/s of the model’s reference server on Decision-Index-style traffic, with the same benchmark scores within noise. In shared-state mode (experimental), requests with several questions about one state reach about 6.0x on the same traffic with no measurable Decision Index change; it is an approximation (about 2 % of multi-question top-1 answers differ). Single-question requests stay at about 3.1x. The code is public under the MIT License. Mining and validating on subnet 92 are always free.

How REAX works →Source code on GitHub →

Two approaches

Choose the approach that fits your data.

Same model profiles and API request shape. Implementations depend on available capacity. Choose per API key or per request with one header.

PEER-TO-PEERElastic capacity

Peer-to-Peer — capacity that scales with your workload

Built to scale worldwide. Peer-to-Peer starts on spare EU capacity. When it is busy, keys you opted in can draw on rented worldwide GPU capacity from independent hardware providers in various countries, including outside the EU/EEA.

Today

EU requests have priority. Enable “Allow worldwide processing” for each key that may leave the EU. Without opt-in, Peer-to-Peer stays on EU capacity and may be queued or rejected when busy. Peer-to-Peer capacity depends on how many network providers are live, so short bursts can get a retryable 503; for steady high volume use an EU key. Existing customers keep EU-only processing unless they opt in. Our gateway does not log payloads; on keys with the separate network-processing opt-in, independent providers of the peer-to-peer network may store content.

  • s1-fast $0.019 · s1-vision $0.031 · s1-pro $0.080 per 1M input tokens
  • Output tokens free
  • Suited to high volume and non-sensitive data
Peer-to-Peer pricing →
EUFlag of the European UnionHighest data protection

EU — for the strictest data-protection requirements

For teams with the highest data-protection demands or strict EU regulatory requirements. EU requests run only on contracted operators inside the EU, and inference data never leaves the EU. EU requests have priority, even if a key allows worldwide Peer-to-Peer processing.

  • s1-fast $0.021 · s1-vision $0.032 · s1-pro $0.084 per 1M input tokens
  • Inference data stays in the EU
  • Contracted EU operators only — EU requests have priority
  • Data Processing Agreement at signup, published sub-processor list
  • Our gateway does not log prompts or answers, and we do not use them for training
EU site →

Sign-in (Google) and payments (Stripe) are handled by the sub-processors listed on our sub-processor page; they never receive your prompts. Worldwide Peer-to-Peer capacity is currently rented through Lium (lium.io), whose independent GPU providers are listed there too. Network processing requires its own per-key opt-in; independent providers may store content.

Built for teams

Everything you need to integrate.

START HERE

Documentation

Quickstarts, request shapes, errors and a migration guide from Jev.

Browse docs →

DATA HANDLING

Trust center

Where inference runs, how data is handled and which contracts you get.

Read trust overview →

LLMS + MCP

For agents

Machine-readable guides and a hosted MCP server.

Agent access →

FAQ

Common questions

Is Decision Models the Jev service?

No. Decision Models is a separate service hosting open-weight decision models behind a Jev-compatible API. Jev is a trademark of TypeSafe AI; references are nominative only. Decision Models is not affiliated with TypeSafe.

How do I get API access?

Sign in with a one-time email link or with Google, create an API key in the dashboard and send your first request. Free allowance by invitation. Top up from EUR 1. Private individuals and businesses worldwide can sign up and top up. Taxes are shown in checkout before you pay.

Is there automatic top-up?

Optional: in the dashboard you can let us charge your saved card when your balance falls below a threshold you set (amount from EUR 10, monthly cap you set, off by default, switch off any time). Each charge gets a normal invoice with VAT.

Are prompts or answers stored?

Our gateway does not log prompts or answers, and we do not use them for training. Independent providers on network-enabled keys may store content; use the EU tier for personal data, secrets or confidential content. For billing we keep usage records such as token counts.

How do Peer-to-Peer and EU differ?

Peer-to-Peer starts on spare EU capacity, with EU requests taking priority. When busy, Peer-to-Peer can draw on rented worldwide GPU capacity from independent hardware providers in various countries, including outside the EU/EEA. This happens only for keys with “Allow worldwide processing” enabled. Without opt-in, Peer-to-Peer stays on EU capacity and may be queued or rejected when busy. Existing customers keep EU-only processing unless they opt in. Our gateway does not log payloads; on keys with the separate network-processing opt-in, independent providers of the peer-to-peer network may store content. Requests to models served by the network (currently: s1-fast-1) go to REAX miners with a track record as the primary route, with our GPUs as fallback; answers are checked on a sampled basis. All other models in this tier run on our GPUs (EU, and with worldwide processing enabled also non-EU partner GPUs). If the network has no capacity for an exact network version, the request is answered on our own GPUs with the same exact model version, never with another model. A retryable 503 is returned only if neither the network nor our own GPUs have capacity. This applies to keys with network processing turned on, which is preselected when you create a Peer-to-Peer key in the dashboard ("Our GPUs only" is selectable) and separate from worldwide processing; existing keys are not changed. The EU tier never uses the network. EU-tier requests always stay in the EU. Processing locations and transfer conditions →

What does worldwide processing allow?

Only Peer-to-Peer requests from keys you opted in may use independent compute providers outside the EU/EEA when spare EU capacity is busy. EU-tier requests always stay in the EU. The switch does not establish a legal basis for transferring personal data: do not send personal data on opted-in keys unless an adequate transfer basis applies to your use (never on network-enabled keys). Read the DPA and sub-processor list.

Can I send images?

Yes, with s1-vision: one image per request (PNG, JPEG or WebP). Text-only requests are billed at the s1-vision text rate; a request that contains an image is billed at the s1-vision image rate for all its input tokens (text and image positions), because image processing takes about 3× the GPU time per token. An image counts as the input tokens the model actually reads.

Is a decision model as accurate as a frontier LLM?

Not for every task. Decision models suit narrow, well-defined decisions. Validate your own examples; use a frontier LLM or human review for open-ended writing and difficult tasks.