Always Updating Lists
•
October 9, 2026
•Updated October 10, 2026
•
8 min read
Every Decision Model on OpenRouter: IDs, Pricing and API Differences
A dated directory of 19 decision-model entries, including Clef, Jev, Mercury Decide, Span-01, Tev1, and the newest Microsoft and Nace AI models.
Co-Founder & Lead Programmer of AcceleratedLogic AI
Catalog checked October 10, 2026. OpenRouter currently lists 19 entries with the
decisions output modality. This directory includes versioned models, free routes, and the moving Jev alias. It covers the category’s catalog snapshot on this date; it is not a claim to list every decision model available anywhere.October 10 update
Re-fetching the public catalog found no additions or changes to the listed models’ input/output rates or context values compared with the October 9 snapshot. This refresh adds the API’s cache-read pricing to the comparison. “Not reported” means the field is absent; it does not mean caching is free or supported.
What is a decision model?
A decision model evaluates supplied state against a question or rubric and returns a result that software can use: a category, a probability, or a score. A support system might select the team that should receive a ticket; a monitoring service might estimate whether a conversation contains a specified behavior. Many models in this category produce probabilities directly instead of generating an explanation.
The common System One primitives are Choice for selecting among named options, Noul for estimating whether a condition holds, and Score for evaluating an ordered rubric. The model and endpoint determine which primitives and inputs are available. OpenRouter’s decision-model overview
Complete OpenRouter decision-model list
The table records exact API context values rather than rounded labels such as “66K.” Input and output prices are USD per one million tokens, taken from the public model catalog. These are a dated snapshot of catalog prices; provider routing, additional services, and promotions can affect a request’s final bill. Each model name links to its OpenRouter listing; the moving alias links to the category catalog.
| Model | Exact OpenRouter ID | Inputs | Catalog context (tokens) | Input / 1M | Output / 1M | Cache read / 1M |
|---|---|---|---|---|---|---|
| TypeSafe: Jev Latest |
~typesafe/jev-latest
|
text | 64,000 | $0.042 | $0 | Not reported |
| Cloudflare: Clef |
cloudflare/clef
|
text, image | 65,536 | $0.24 | $0 | Not reported |
| Cloudflare: Clef Flash |
cloudflare/clef-flash
|
text, image | 65,536 | $0.038 | $0 | Not reported |
| Cloudflare: Clef Omni |
cloudflare/clef-omni
|
text, image | 65,536 | $0.15 | $0 | Not reported |
| Inception: Mercury Decide |
inception/mercury-decide
|
text | 65,536 | $0.02 | $0 | $0 |
| Inception: Mercury Decide (free) |
inception/mercury-decide:free
|
text | 65,536 | $0 | $0 | Not reported |
| Jared Palmer: Kev 4B |
jaredpalmer/kev-4b
|
text | 8,192 | $0.042 | $0 | Not reported |
| LiquidAI: d1 |
liquid/d1
|
text | 65,536 | $0.04 | $0 | $0.04 |
| Microsoft: Microsoft-Decision-1 |
microsoft/microsoft-decision-1
|
text | 32,768 | $0.042 | $0 | Not reported |
| Nace.AI: Drex v1.5 |
nace-ai/drex-v1.5
|
text | 131,072 | $0.04 | $0 | Not reported |
| OpenAI: GPT-6 Luna Decisions |
openai/gpt-6-luna-decisions
|
text, image | 1,050,000 | $0.1 | $0 | $0 |
| Perplexity: Decider V1.1 27B |
perplexity/pplx-decider-v1.1-27b
|
text, image | 262,144 | $0.02 | $0 | Not reported |
| Respan: Span-01 |
respan/span-01
|
text | Not reported | $0.02 | $0 | Not reported |
| Respan: Span-01 Lite |
respan/span-01-lite
|
text | Not reported | $0 | $0 | Not reported |
| Respan: Span-01 Lite (free) |
respan/span-01-lite:free
|
text | Not reported | $0 | $0 | Not reported |
| Together: Tev1 4B Experimental |
togethercomputer/tev1-4b-experimental
|
text | 32,768 | $0.042 | $0 | Not reported |
| TypeSafe: Jev 1.13 |
typesafe/jev-1.13
|
text | 64,000 | $0.042 | $0 | Not reported |
| Upstage: Solar Decide |
upstage/solar-decide
|
text | 524,288 | $0.05 | $0 | $0.05 |
| Upstage: Solar Decide Flash |
upstage/solar-decide-flash
|
text | 524,288 | $0.05 | $0 | $0.05 |
Source: OpenRouter Models API filtered to decisions. All 19 entries currently report $0/M output pricing. A paid input rate still produces a bill even when the output rate is zero. Respan’s context value is zero in the API, so its rows say “Not reported”; zero does not establish an unlimited or unusable context window.
Cloudflare Clef, Clef Flash, and Clef Omni
Clef Flash is the 9B member, Clef is the 27B member, and Clef Omni is the mixture-of-experts member of the family. OpenRouter currently marks all three as accepting text and images. Their catalog input rates are $0.038, $0.24, and $0.15 per million tokens respectively. Clef Flash, Clef, Clef Omni
Important serving limit: the Clef and Clef Flash cards warn that Workers AI reads roughly the first 2,000 tokens of long text state, with images accounted for separately. Their advertised 65,536-token catalog context is therefore not a promise that the Workers AI route reads an entire long document. Check the chosen provider’s limits and put decisive evidence inside the portion it actually receives.
TypeSafe Jev and the moving alias
Jev 1.13 is the versioned TypeSafe entry.
~typesafe/jev-latest follows the latest Jev release. An alias is convenient for receiving updates; the versioned ID makes an evaluation easier to reproduce. The current catalog reports 64,000 tokens for both. One OpenRouter overview page still describes an older 32,000-token limit, which is why this table explicitly uses the dated API values. Jev model cardJev Router is a separate text-output routing product and is not included in this decisions-only list. Our Jev Router guide explains its downstream model selection and dynamic pricing.
Together Tev1: an API exception worth knowing
Tev1 4B Experimental appears in the decision category, but its card specifies ordinary Chat Completions for inference. It selects a letter from 2–24 labeled options, and application code maps that letter back to an option key. The card recommends temperature 0, a small eight-token output cap, and disabled thinking. Use the documented Chat Completions payload for this model rather than assuming the System One schema. Tev1 card and calling instructions
Respan Span-01 and the free Lite routes
Span-01 evaluates behaviors in conversation spans. Span-01 Lite is the lighter tier, and the API lists both
respan/span-01-lite and respan/span-01-lite:free. Both currently have zero input and output prices. Keep the exact route ID in your records instead of treating every duplicate-looking label as a separate trained checkpoint. Span-01, Span-01 Lite, Lite free routeInception Mercury Decide and Upstage Solar Decide
Mercury Decide has a paid route and a separate
:free route. Its System One interface exposes typed decisions, and the catalog currently prices the paid route at $0.02/M input. Mercury DecideSolar Decide and Solar Decide Flash are Upstage’s entries. Both expose a 524,288-token catalog context and currently cost $0.05/M input. The Flash card describes it as the faster variant; that is a product claim, not an independent speed result from this article. Solar Decide, Solar Decide Flash
Microsoft, Nace AI, Liquid AI, Jared Palmer, OpenAI, and Perplexity
Microsoft-Decision-1 is a compact option-scoring model whose card describes continuing weight updates. Its API ID alone should not be treated as proof of a permanently frozen checkpoint. Drex v1.5 from Nace AI provides a 131,072-token catalog context and a System One interface. Microsoft-Decision-1, Drex v1.5
Liquid AI d1 provides another typed decision endpoint. Jared Palmer’s Kev 4B is an open-weight model with a System One contract and an 8,192-token catalog context; its card links the code, model cards, and evaluations. d1, Kev 4B
GPT-6 Luna Decisions accepts text and images and has the largest context value in this snapshot: 1,050,000 tokens. Perplexity Decider V1.1 27B also accepts text and images, with 262,144 tokens. The older Perplexity V1 listing appears in usage rankings, but the current decisions-filtered API response contains V1.1; this table follows the live catalog. GPT-6 Luna Decisions, Decider V1.1
Calling and discovering decision models
The default OpenRouter Models API response contains text-output models. To obtain this category, request
output_modalities=decisions; output_modalities=all also includes other output types. Models API filtering documentationcurl "https://openrouter.ai/api/v1/models?output_modalities=decisions"OpenRouter documents
POST https://openrouter.ai/api/alpha/decisions for its Decisions API and POST https://openrouter.ai/api/v1/systemone for the System One surface. Read the chosen model’s instructions for the supported questions, state format, and response fields. Tev1’s documented Chat Completions behavior is the exception described above. A generic text-chat client does not automatically implement these decision-specific payloads. Decisions request referenceHow to compare these models usefully
Start with the decision your software actually needs to make, define the allowed answers, and test on labeled examples from that workflow. Record classification accuracy, missed cases, probability calibration, latency, and total cost. A confidence value needs validation on your data before it becomes an automatic-action threshold.
For example, a ticket-routing evaluation should include clear requests, mixed-topic tickets, and requests that belong to none of the supplied teams. A behavior-scoring evaluation should include both obvious violations and ambiguous cases. Keep the decision labels, thresholds, route IDs, and provider configuration fixed when comparing runs.
OpenRouter’s decision rankings show adoption by request volume. They help identify popular routes, but request volume alone does not establish classification accuracy for a particular application. This article is a catalog comparison, with original integration notes; it does not claim independent accuracy or latency measurements.
Which entries are free?
This snapshot contains three entries with zero input and output prices: Mercury Decide (free), Span-01 Lite, and Span-01 Lite (free). Availability, quotas, and access terms still come from the selected route. Recheck the linked model card before relying on a free route in a workflow. The other entries have a paid input rate, even though their output columns show zero.
Read Next
AI Models
•
3 min read
Crouton Alpha HTML Coding Check: Glass Crane, Globe & Wizard Game
A short creative-coding run with the Crouton Alpha label in OpenRouter: a glass-crane SVG, a procedural Three.js globe, and a wizard-island game, with original HTML and clearly scoped observations.
Mohid Mirza
AI Models
•
10 min read
Recursive Flint: All Four LogicMark Benchmarks, Scores and Failures
We test Recursive Flint on both LogicMark 1.0.0 suites in Tool-assisted and Quick modes, documenting objective scores, protocol failures, timing and API usage.
Mohid Mirza
AI Models
•
8 min read
MiniMax M3.1 Flash: Four HTML Coding Benchmarks on Free OpenCode
MiniMax M3.1 Flash tested through free OpenCode: original crane, globe, wizard-game and task-board HTML, browser screenshots, timings, downloadable evidence, 16 passing board checks and a reproduced game physics failure.
Mohid Mirza