If your decisions read customer or company data, Mouse-1 is the safer default: zero data retention on every request, a fitted confidence cutoff that tells you when not to trust an answer, and tuning on your own outcomes. If you need the cheapest per-token price or long inputs, Perplexity Decider v1.2 wins: $0.02 per 1M tokens, 256K context and higher public benchmark scores.
Perplexity's Decider API (POST /v1/decisions) serves Decider v1.2, a 27B model, and Decider Flash v1.2, a 4B model. Requests use a state plus named noul, choice and score questions, the same shape Mouse-1 uses. The difference is less about the model than about what each platform does with your data and how much control it gives you over the answers.
At a glance
| Mouse-1 | Perplexity Decider v1.2 | |
|---|---|---|
| Zero data retention | Default on every request | Eligible customers, written agreement |
| Calibrated abstention | /v1/calibrations + abstain flag | Pick thresholds yourself |
| Fine-tuning on your decisions | /v1/tunings | Not offered |
| Claude Code tool-call gate | /v1/safeguards | Not offered |
| Prompt-injection results | Published: 0% flipped | None published |
| Public tool benchmarks (avg of 4) | 84.9% | 87.6% |
| Price per 1M input tokens | $0.10 | $0.02 (Flash: $0.01) |
| Context | 16,384 state tokens | 256K |
| Images | 8 per request (beta) | 8 per question |
| Rate limit | 10 RPS | 10 RPS |
Data retention is the deciding line
A decision model sees the most sensitive part of an agent's work: the diff it wants to ship, the email it wants to send, the customer record it is about to change. Mouse-1 sends Mouse-Data-Retention: none on every response and never logs states, questions or answers unless you explicitly turn on stored decisions for tuning. Perplexity's docs say zero data retention "is not default" and is available only to eligible customers through an account-level agreement.
Control over the answers
- Calibration.
POST /v1/calibrationsfits a cutoff on 200–400 of your labelled examples so each answer carriesmouse.abstain, meaning "send this one to a human". Perplexity recommends choosing thresholds on your own data, but has no endpoint for it. - Tuning. Mouse-1 can be trained on outcomes you post back, and each tuning is tested against the base model before you can use it. Perplexity's Decider docs list no fine-tuning.
- Injection. On our planted-instruction test, none of Mouse-1's decisions changed. Perplexity publishes no injection results for Decider.
Public tool benchmarks: Perplexity leads
Average of BFCL, API-Bank, When2Call and PhishNChips (%) · Higher is better
| Perplexity Decider v1.2 | 87.6% |
|---|---|
| Mouse-1 | 84.9% |
Function calling, exact match (%) · Higher is better
| Perplexity Decider v1.2 | 97.8% |
|---|---|
| Mouse-1 | 97.6% |
Next API out of 53 (%) · Higher is better
| Perplexity Decider v1.2 | 87.4% |
|---|---|
| Mouse-1 | 83.7% |
Answer, call, ask or decline (%) · Higher is better
| Perplexity Decider v1.2 | 90.3% |
|---|---|
| Mouse-1 | 85.8% |
Phishing or legitimate email (%) · Higher is better
| Perplexity Decider v1.2 | 74.7% |
|---|---|
| Mouse-1 | 72.3% |
Perplexity Decider v1.2 is ahead on API-Bank, When2Call and PhishNChips, and within 0.2 points of Mouse-1 on BFCL. On these four benchmarks it is the stronger model.
How we measured
- Perplexity's scores are from the Jev Decision Index board as of October 20, 2026. Mouse-1 is not on the board; its scores come from our own run of the board's kit on the same rows.
- Product facts are from Perplexity's API docs and pricing page, checked October 20, 2026.
- We have not run Perplexity Decider through our agent decision suites or our injection test.
Sources
Checked October 20, 2026.
- Perplexity Decisions API reference
- Perplexity API pricing
- Perplexity privacy and security
- Jev Decision Index board data
- Mouse-1 API reference and pricing
See also: Mouse-1 vs Fastino GLiDE · Mouse-1 vs OpenAI Decisions · All comparisons
FAQ
Is Perplexity Decider better than Mouse-1?
On the four public tool benchmarks of the Jev Decision Index, Perplexity Decider v1.2 averages 87.5% against Mouse-1's 84.9%, and it costs $0.02 per 1M input tokens against $0.10. Mouse-1 offers zero data retention on every request, a calibration endpoint and fine-tuning, none of which Perplexity's Decider API documents.
Does Perplexity Decider have zero data retention?
Perplexity's docs say zero data retention is not the default and is available only to eligible customers under a written agreement. Mouse-1 applies zero data retention to every request by default.
Can I fine-tune Perplexity Decider?
Perplexity's Decider docs do not offer fine-tuning or a calibration endpoint; they suggest picking thresholds on your own labelled data. Mouse-1 has /v1/calibrations for a fitted confidence cutoff and /v1/tunings to train on your decision outcomes.
Which has longer context, Mouse-1 or Perplexity Decider?
Perplexity Decider: the state plus one question can be up to 256K tokens. Mouse-1 reads up to 16,384 state tokens.