System One models come in two classes. A decision model reads content you supply, answers typed questions about it, and returns calibrated probabilities instead of prose; Jev was the first. An extraction model, a class Cloudflare introduced with Clairic, reads typed values such as amounts and dates straight from the input. Extraction models are listed in their own section. What a System One model is, what a decision model is, and what an extraction model is.
elephant-1 is a 9B decision model from Base Labs, Baseten's research team, post-trained from Qwen3.5-9B. It answers typed Choice, Score and Noul questions about text and images with a probability for every option, runs on Baseten's Model APIs and OpenRouter, and has MIT weights on Hugging Face.
mammoth-1 is a 27B decision model from Base Labs, Baseten's research team, post-trained from Qwen3.5-27B. It answers typed Choice, Score and Noul questions about text and images with a probability for every option, runs on Baseten's Model APIs and OpenRouter, and has MIT weights on Hugging Face.
d1-ultra is Liquid AI's larger hosted decision model, released on 15 October 2026. It answers typed Choice, Score and Noul questions about text and images with probabilities and writes no prose. It costs $0.042 per million input tokens and leads the decision models on the Artificial Analysis Decision Index at 65.2.
Span-02 is Respan's decision model, released on 15 October 2026. It answers typed Choice, Score, Noul and number questions about text, JSON and images, can think briefly when unsure within a time budget you set, and costs $0.01 per million input tokens. Its weights, a Gemma 4 26B-A4B fine-tune, are on Hugging Face.
SubQ Snap is Subquadratic's hosted System One model. It answers typed Noul, Choice and Score questions about text or images with probabilities, reads up to 384K tokens in one request and costs $0.04 per million input tokens. The weights are closed, and the API is in public preview.
Clef-omni is Cloudflare's multimodal decision model. It reads audio and video as well as text and images, answers the same typed questions as Jev, and runs on Workers AI at $0.15 per million input tokens. It is a 30B mixture-of-experts model with 3B parameters active, and the weights are open under Apache 2.0. Cloudflare launched it on 9 October 2026.
Microsoft-Decision-1 is Microsoft's System One model, post-trained from Qwen3.5-9B. It returns a calibrated probability for each fixed answer option to yes/no, multiple-choice and rating questions, through Microsoft Foundry at $0.042 per million input tokens. The weights are closed.
Celeris-1 Decision is a hosted System One model from Celeris. It answers typed Noul, Choice and Score questions about text, structured data or images with probabilities, and can add a one-sentence reason per answer. It costs $0.04 per million input tokens, and the weights are closed.
Clef is Cloudflare's own decision model, in two sizes: Clef at 27B and Clef-flash at 9B. Both run on Workers AI, take the same state and typed questions as Jev, and read images as well as text. Clef has a 65,536-token context on Workers AI and Clef-flash 24,576. The weights are open under Apache 2.0. Cloudflare launched them on 1 October 2026.
pplx-decider is Perplexity's multimodal System One model, in two sizes: pplx-decider-v1.2 at 27B and pplx-decider-flash-v1.2 at 4B. Both answer typed Choice, Score and Noul questions about text or images with probabilities, through the Decisions API at $0.02 and $0.01 per million input tokens. The weights are on Hugging Face under Apache 2.0.
GLiDE is Fastino's decision model. Version 1.2, released on 22 October 2026, answers typed Noul, Choice and Score questions about text, JSON or images in one forward pass and returns a probability for every option, at $0.04 per million input tokens. Its weights, post-trained from Qwen3.8-27B, are on Hugging Face under Fastino's own licence, which limits production use by large companies.
Mercury Decide is Inception's decision model, served as a System One endpoint. It answers typed choice, score and yes/no questions with a probability and writes no prose. It costs $0.02 per million input tokens on Inception's API and OpenRouter, and a free tier stays on OpenRouter.
d1 is Liquid AI's decision model family. It answers typed Choice, Score and Noul questions with probabilities and writes no prose. The hosted d1 launched on Liquid's API as d1:free on 29 September 2026, and the paid d1 costs $0.02 per million input tokens. Liquid added two open-weight versions, d1-3B and d1-omni-600M, on 7 October 2026.
OpenAI's Decisions API is a hosted System One model built on GPT-6 Luna. You send text or images and a question with fixed answers, and it picks one. It has been in public beta since 6 October 2026 at $0.10 per million input tokens, with free output.
Kai 1 is Hanzo AI's System One model. It answers typed Choice, Score and Noul questions with calibrated probabilities through the Hanzo API at $0.021 per million input tokens, and writes no prose. The weights are closed and the model is API only.
Drex is Nace.AI's family of decision models. Drex Max, the most accurate, reads text and images. Drex 2.0, the default, is a 35B mixture of experts with open weights. Both answer typed Choice, Score and Noul questions with a probability for every option through Nace's own /v1/systemone API at $0.06 per million input tokens. Drex 1.5, a 9B model, stays on sale at $0.05.
Span-01 is Respan's behavior classifier for agent traces. It reads a conversation span and behaviors you define in plain language, and returns the probability that each is present, absent or not observable. It costs $0.02 per million input tokens, and a free Lite tier exists.
Tev1-4B-experimental is Together AI's Jev-like classifier, a Qwen3.5-4B fine-tune that reads a state and a set of 2 to 24 options and returns one answer letter. Together released it alongside the training recipe and a tutorial on 23 September 2026.
Decider 1 is meraGPT's System One model. It answers typed Choice, Score and Noul questions with probabilities over a 4,096-token request and writes no prose. It has been on meraGPT's hosted API since 22 September 2026, with no public weights.
Solar Decide is Upstage's System One model, served on Solar Mini 4 with a 512K-token context. It answers typed Choice, Score and Noul questions with probabilities, writes no prose, and has been in beta on Upstage's API since 22 September 2026.
Jev 1.13 is the first System One model. It reads text you supply, answers typed questions about it with calibrated probabilities, and returns no generated prose. TypeSafe AI shipped it on 15 September 2026 as a closed managed API.
These are published weights you download and run on your own hardware, with no bill beyond your own GPUs. A few also have a paid hosted endpoint, which each model page lists. A company or a single developer may have made them. They are listed once they clear the bar below.
H2O-Lightning-31B is an open-weights decision model from H2O.ai, a fine-tune of Google's Gemma-4-31B-it. It answers typed Choice, Score and Noul questions about text and images with a probability for every option. The weights are gated on Hugging Face under H2O's non-commercial licence, which allows non-commercial, non-production use only.
Jeffan-31b-1710 is an open-weights decision model by Trong Vu, a full fine-tune of Google's Gemma-4-31B-it published on Hugging Face on 19 October 2026. It reads one multiple-choice question at a time about a state and returns a probability for every option. Weights and inference code are Apache 2.0.
Quyet-1.0-Large is an open-weights decision model by Chinh Nguyen: Gemma-4-31B-it with a merged LoRA, Apache 2.0. It answers typed Choice, Score and Noul questions with calibrated probabilities in one pass, and leads JevBench's Capability ranking.
Torchcast Decision 27B is an open-weights decision model from Torchcast AI, a Singapore forecasting company. It answers typed choice, noul and score questions with a probability for every option, accepts up to 262,144 tokens of state, and was published on Hugging Face on 3 October 2026 under a non-commercial CC BY-NC 4.0 licence.
H2O-Lightning-4B is an open-weights decision model from H2O.ai, a fine-tune of Qwen3.5-4B under Apache 2.0. It answers typed Choice, Score and Noul questions about text and images, returns a probability for every option, and tops JevBench's official composite on the open-weights board.
Strands-Decider 2B is a small open-weights decision model from Strands Labs, the experimental arm of the Strands Agents project, which is linked to AWS. A LoRA adapter and a small head on Qwen3.5-2B-Base answer typed Noul, Choice and Score questions with a confidence on each, and it runs on your own GPU, Mac or CPU under Apache 2.0.
GLiNER2.5-Decide is Fastino Labs' open-weights System One model. A 340M-parameter encoder answers typed classification questions, extracts spans and relations, and enforces cross-decision rules, and Fastino shipped it on 24 September 2026 under Apache 2.0.
CLM (Contrastive Language Models) is an open-weights System One model that scores a state against a set of candidate actions with a contrastive objective instead of generating text. CLM-8B serves a TypeSafe-compatible API and Jacky Kwok's team released it, its data recipe and its scaling-law study on 23 September 2026.
Surogate Rune is an open-weights decision model from Surogate, a full fine-tune of Gemma 4 26B-A4B that reads text and images. It answers typed Choice, Score and Noul questions with a probability for each option, and you download the weights from Hugging Face after accepting the gate terms.
Laya is an open-weights System One model from Convai Innovations. It answers typed Choice, Score and Noul questions about text you supply, returns probabilities instead of prose, and runs on your own GPU or CPU under Apache 2.0.
Von is an archived open-weights decision model by the solo developer wfzyx. A 395M ModernBERT-large encoder answers typed Choice, Noul and Score questions in one forward pass and serves the same /v1/systemone API as Jev, under Apache 2.0. Its author stopped maintaining it in October 2026 and points new users to Liquid AI's d1-omni-600M.
Kev is an open-source family of System One models from Jared Palmer, at 0.8B, 4B, 9B and 27B on Qwen3.5 and Qwen3.8 bases. It answers typed Choice, Score and Noul questions with probabilities, serves TypeSafe's /v1/systemone API, and runs on your own GPU under Apache 2.0, with the 0.8B to 9B models also on a Mac. Kev-4B is also on OpenRouter at $0.042 per million input tokens.
Cloudflare calls these a second System One class. An extraction model returns typed values read from the input, computed from it or predicted beyond it, such as a total, a date or a forecast, instead of picking from options you list. They are listed here and kept out of the decision-model tables and the ranking. What an extraction model is.
Clairic is Cloudflare's first extraction model, in preview on Workers AI. It reads text and up to four images and returns typed values (spans, numbers, dates, durations, lists, estimates and forecasts) with probabilities, in one pass and with no output tokens. It costs $0.10 per million input tokens. Cloudflare launched it on 19 October 2026.
The headline numbers from each model page. For the full side-by-side, with benchmarks, limits and SDKs, go to the compare hub, or open one of its head-to-head pages.
| Model | Vendor | Status | Access | Question types | Input price | Context | Latency | Licence |
|---|---|---|---|---|---|---|---|---|
| elephant-1 | Baseten | Generally available | Hosted API | Choice, Score, Noul | Not published as of 23 Oct 2026 | 16,384 tokens | Not published as of 23 Oct 2026 | MIT |
| mammoth-1 | Baseten | Generally available | Hosted API | Choice, Score, Noul | Not published as of 23 Oct 2026 | 16,384 tokens | Not published as of 23 Oct 2026 | MIT |
| d1-ultra | Liquid AI | Generally available | Hosted API | Choice, Score, Noul | $0.042/MTok | 32,768 tokens | Not published as of 23 Oct 2026 | Not published as of 23 Oct 2026 |
| Span-02 | Respan | Generally available | Hosted API | Choice, Score, Noul | $0.01/MTok | 131,072 tokens | 33 to 822 ms4 | Apache-2.0 |
| SubQ Snap | Subquadratic | Early access | Hosted API | Choice, Score, Noul | $0.04/MTok | 393,216 tokens | Not published as of 23 Oct 2026 | Not published as of 23 Oct 2026 |
| Clef-omni | Cloudflare | Generally available | Hosted API | Choice, Score, Noul | $0.15/MTok | 64,000 tokens | 130 to 150 ms6 | Apache-2.0 |
| Microsoft-Decision-1 | Microsoft | Generally available | Hosted API | Choice, Score, Noul | $0.042/MTok | 32,768 tokens | 85 to 125 ms7 | Not published as of 23 Oct 2026 |
| Celeris-1 Decision | Celeris | Generally available | Hosted API | Choice, Score, Noul | $0.04/MTok | Not published as of 23 Oct 2026 | 67 to 81 ms8 | Not published as of 23 Oct 2026 |
| Clef | Cloudflare | Generally available | Hosted API | Choice, Score, Noul | $0.24/MTok | 65,536 tokens | 38.8 to 209.3 ms9 | Apache-2.0 |
| pplx-decider | Perplexity | Generally available | Hosted API | Choice, Score, Noul | $0.02/MTok | 262,144 tokens | Not published as of 23 Oct 2026 | Apache-2.0 |
| GLiDE | Fastino Labs | Generally available | Hosted API | Choice, Score, Noul | $0.04/MTok | 65,536 tokens | 360 to 516 ms11 | Fastino Community License |
| Mercury Decide | Inception | Generally available | Hosted API | Choice, Score, Noul | $0.02/MTok | 65,536 tokens | Not published as of 23 Oct 2026 | Not published as of 23 Oct 2026 |
| d1 | Liquid AI | Generally available | Hosted API | Choice, Score, Noul | $0.02/MTok | 65,536 tokens | Not published as of 23 Oct 2026 | LFM Open License v1.0 (open d1-3B and d1-omni-600M; the hosted d1 has no published licence) |
| OpenAI Decisions API | OpenAI | Early access | Hosted API | Choice, Score, Noul | $0.1/MTok | Not published as of 23 Oct 2026 | Not published as of 23 Oct 2026 | Not published as of 23 Oct 2026 |
| Kai | Hanzo AI | Generally available | Hosted API | Choice, Score, Noul | $0.021/MTok | 1,024 tokens | Not published as of 12 Oct 2026 | Not published as of 12 Oct 2026 |
| Drex | Nace.AI | Generally available | Hosted API | Choice, Score, Noul | $0.06/MTok | 262,144 tokens | Not published as of 23 Oct 2026 | Nace.AI Open RAIL-M |
| Span-01 | Respan | Generally available | Hosted API | Noul | $0.02/MTok | Not published as of 23 Oct 2026 | Not published as of 23 Oct 2026 | Not published as of 23 Oct 2026 |
| Tev1 | Together AI | Early access | Hosted API | Choice | $0.042/MTok | 2,048 tokens | Not published as of 23 Oct 2026 | Not published as of 23 Oct 2026 |
| Decider 1 | meraGPT | Generally available | Hosted API | Choice, Score, Noul | $0.03/MTok | 4,096 tokens | Not published as of 12 Oct 2026 | Not published as of 12 Oct 2026 |
| Solar Decide | Upstage | Early access | Hosted API | Choice, Score, Noul | $0.1/MTok | 524,288 tokens | Not published as of 9 Oct 2026 | Not published as of 9 Oct 2026 |
| Jev | TypeSafe AI | Generally available | Hosted API | Choice, Score, Noul | $0.042/MTok | 64,000 tokens | 70 to 500 ms21 | Not published as of 23 Oct 2026 |
| Model | Vendor | Status | Access | Question types | Input price | Context | Latency | Licence |
|---|---|---|---|---|---|---|---|---|
| H2O-Lightning-31B | H2O.ai | Generally available | Open weights, self-hosted | Choice, Score, Noul | No hosted API | 131,072 tokens | 46 to 57 ms1 | H2O Model Non-Commercial Licence v1.1 |
| Jeffan-31b-1710 | Trong Vu | Generally available | Open weights, self-hosted | Choice | No hosted API | 8,192 tokens | 43 to 121 ms2 | Apache-2.0 |
| Quyet-1.0-Large | Quyet | Generally available | Open weights, self-hosted | Choice, Score, Noul | No hosted API | 8,000 tokens | 105.4 to 115.8 ms3 | Apache-2.0 |
| Torchcast Decision 27B | Torchcast AI | Generally available | Open weights, self-hosted | Choice, Score, Noul | No hosted API | 262,144 tokens | 33 to 49 ms4 | CC-BY-NC-4.0 |
| H2O-Lightning-4B | H2O.ai | Generally available | Open weights, self-hosted | Choice, Score, Noul | No hosted API | 40,960 tokens | 28.8 to 30.2 ms5 | Apache-2.0 |
| Strands-Decider 2B | Strands Labs | Generally available | Open weights, self-hosted | Noul, Choice, Score | No hosted API | 4,096 tokens | Not published as of 23 Oct 2026 | Apache-2.0 |
| GLiNER2.5-Decide | Fastino Labs | Generally available | Open weights, self-hosted | Choice, Score, Noul | No hosted API | Not published as of 23 Oct 2026 | 38 to 167 ms7 | Apache-2.0 |
| CLM | Contrastive-LM | Generally available | Open weights, self-hosted | Choice, Score, Noul | No hosted API | 2,048 tokens | Not published as of 23 Oct 2026 | Apache-2.0 |
| Surogate Rune 26B-A4B | Surogate | Generally available | Open weights, self-hosted | Choice, Score, Noul | No hosted API | Not published as of 23 Oct 2026 | Not published as of 23 Oct 2026 | Apache-2.0 |
| Laya | Convai Innovations | Generally available | Open weights, self-hosted | Choice, Score, Noul | No hosted API | 512 tokens | 32.8 to 39.5 ms10 | Apache-2.0 |
| Von | wfzyx | Deprecated | Open weights, self-hosted | Choice, Noul, Score | No hosted API | 8,192 tokens | Not published as of 23 Oct 2026 | Apache-2.0 |
| Kev | Jared Palmer | Generally available | Open weights, also hosted on OpenRouter | Choice, Score, Noul | $0.042/MTok | 8,192 tokens | 41.5 to 145.2 ms12 | Apache-2.0 |
| Model | Vendor | Status | Access | Field types | Input price | Context | Latency | Licence |
|---|---|---|---|---|---|---|---|---|
| Clairic | Cloudflare | Early access | Hosted API | Text, Number, Date, Time, Datetime, Duration, List, Estimate, Forecast | $0.1/MTok | 65,536 tokens | Not published as of 20 Oct 2026 | Not published as of 20 Oct 2026 |
There is no single answer, because the vendors' own benchmark tables each pick different tasks and run the rivals themselves. The fair comparison is an independent score on the same suite, such as the Artificial Analysis Decision Index, the Jev Decision Index or JevBench, which 29 of the 33 models here carry on their pages, read alongside the price, context window and latency you need. The ranking of the best AI decision models combines the three boards into one System One Score, and the compare hub puts every model in one table and pairs them head to head.
Close, but not the same. A classification model is trained for one fixed label set and returns one label. A decision model reads the content you supply at request time and answers typed questions you write for it, Choice, Score or Noul, with a calibrated probability for every option, so the label set changes per request and no retraining is involved. "Decision model" is the term for the product; "classification model" describes what it does to one question.
H2O-Lightning-31B, Jeffan-31b-1710, Quyet-1.0-Large, Torchcast Decision 27B, H2O-Lightning-4B, Strands-Decider 2B, GLiNER2.5-Decide, CLM, Surogate Rune 26B-A4B, Laya, Von and Kev: 12 of the 33 models listed publish weights you download and run yourself. The rest are hosted APIs only. An open-source model is listed once its repository has 1,000 or more stars, its weights are on Hugging Face, it has had a commit in the last 30 days and it publishes its own evaluation results.
Hosted models charge for input tokens only. Published prices run from $0.01 per million input tokens (Span-02) to $0.24 (Clef), as of 23 Oct 2026. elephant-1 and mammoth-1 have no published price. Open-weight models cost the GPU time you run them on and nothing per token.
A model is listed here when three things are true. Its vendor has announced it publicly. It returns typed answers with probabilities rather than generated text. And every number on its page, from price to rate limit, comes from a public source that the page links.
Open-source models have a higher bar. The GitHub repository needs 1,000 or more stars, the weights must be on Hugging Face, there must be a commit in the last 30 days, and the project must publish its own evaluation results. Projects below that bar go in the alternatives directory instead.
Know a model that fits? Send it in with a link to the announcement.