Tasks and inputs
Reflex is listed for Text classification, Content Moderation, Structured Decisions. It accepts text, image and produces decision. Use the publisher’s input format and preprocessing for your chosen deployment.
Morph Reflex is a family of fast classifier models that label agent turns (jailbreak, looping, frustration and more) with per-class scores.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Unverified: a catalogue lists $0.042 per 1M input tokens and free output; Morph's own pages did not confirm. Check your Morph account.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Agents fail in recognisable ways. They loop, they leak their reasoning, they meet a jailbreak, they frustrate the user. Morph Reflex is built to spot those moments as they happen. Each one is a small, fast classifier. Send a turn; get back a label with a score for every class, not a paragraph of commentary.
Morph ships a default set, each addressed by its own model ID:
| Model ID | Labels |
|---|---|
jailbreak |
benign, jailbreak |
guardrail |
false, true |
leaked-thinking |
clean, leaked |
stuck-in-a-loop |
progressing, looping |
incomplete-thought |
complete, incomplete |
user-frustrated |
Frustrated, Not Frustrated |
ambiguity |
low, med, high |
difficulty |
easy, medium, hard |
domain |
general, summary, coding, design, data |
Labels are case-sensitive. That matters when you wire Reflex into an evaluator that matches category names exactly.
The direct endpoint, /v1/reflex/predict, returns a classes array with every label and score. An OpenAI-compatible base URL also exists. Observability tools can use it to treat any of these classifiers as a categorical judge. The compatible route returns only the top label and no rationale. For multi-label classifiers such as domain, use the direct endpoint if you need every matching label.
You are not stuck with the defaults. Train your own classifier from labelled examples, or let Morph synthesise a dataset from a plain description of what you want caught. Quick.
A model catalogue lists a single "Reflex" entry with a 65,536-token input cap, $0.042 per 1M input tokens, free output, image input and a 27B Qwen 3.5 base. Morph's own pages describe them as text classifiers and did not confirm the price, parameter count, base model or image support when we checked. Release dates also differ by a day (3 or 4 November 2026). Treat those figures as unverified and price a real workload on your Morph account. It is a separate product from Morph's Router.
Maybe. Export a few hundred real agent turns, label them yourself, and run them through the Reflexes you care about. Plot precision and recall at several score thresholds. A jailbreak detector that blocks 3% of legitimate users is a support problem, so pick the threshold that matches your tolerance. Compare it with an open option such as GPT OSS Safeguard 20B, and see how it fits among other structured decision models.
Compare the model on the work you need to complete.
Reflex is listed for Text classification, Content Moderation, Structured Decisions. It accepts text, image and produces decision. Use the publisher’s input format and preprocessing for your chosen deployment.
Documented limits: 65.54K context. The listed route has starting rates of $0.042 input per million tokens. Test labelled examples and choose thresholds from the cost of each error. Validate probabilities against your application’s outcomes.
Use the model identifier and account requirements for the selected API route. Confirm its request limits and output format with one complete task before connecting a larger workflow.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
Gemini Decision 1 powers Google's Decisions API, returning calibrated yes/no, choice and rubric answers instead of text, at $0.10 per 1M input tokens.
Perplexity
pplx-decider-flash-v1.3 is Perplexity's small, half-price decision model, distilled from the 27B decider, for high-volume classification and span extraction.
Perplexity
pplx-decider-v1.3 is Perplexity's 27B-class decision model: it returns probabilities for choice, yes/no, score and new span-locate questions, not text.
Butler
Cicero is Butler Compute's open-weight MoE decision model: up to 64 typed questions per state, a calibrated probability for every option, no text.
Sarvam
Sarvam Prism is a 105B decision model that returns calibrated probabilities for yes/no, choice and score questions in English and 10 Indian languages.
levanto
Levanto Sage 2 is listed as a new version of Levanto's hosted Sage decision model; Levanto's own docs do not yet name a version 2.