# pplx-decider-flash (slim)

> pplx-decider-flash is Perplexity's 4B decision model, a Qwen3.5-4B fine-tune distilled from the 27B pplx-decider. It answers yes or no, choice and score questions about text, JSON and images with probabilities, through the hosted Decisions API or open weights.

- Full: https://www.anchorterminal.com/tools/pplx-decider-flash.md (~12,200 tokens) · this version ~2,180 tokens · JSON https://www.anchorterminal.com/tools/pplx-decider-flash.json · canonical https://www.anchorterminal.com/tools/pplx-decider-flash
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-10-18 · pricing https://www.anchorterminal.com/tools/pplx-decider-flash#pricing · reviews https://www.anchorterminal.com/tools/pplx-decider-flash#reviews
- Licence: CC BY 4.0, credit Anchor Terminal (anchorterminal.com) and link this page when you pass anything on

**B · 67.9/100 · rank #260 of 1133 · #6 in Decision models · not agent-ready · confidence medium**


Assessment: pplx-decider-flash runs on the same hosted Decisions API as the 27B model, with an OpenAPI description, typed questions, Python and TypeScript SDK methods and $0.01 per million input tokens, and its weights are public under Apache-2.0. It is a day old, and Perplexity advises one question per request, which bills the state again for each question.

## Facts

- Kind: Model API · vendor: Perplexity · category: Decision models · legal entity: Perplexity AI, Inc. · provenance 82/100
- Endpoint: `https://api.perplexity.ai/v1/decisions` (HTTP)
- Auth: API key · pricing: Pay per use · x402: no · licence: Apache-2.0 for the published weights and inference code. The hosted Decisions API is a proprietary service
- Probe metrics: not measured yet (probes haven't run)
- Surface graded: Perplexity's hosted Decisions API, `POST /v1/decisions`, with `pplx-decider-flash-v1.2`. The open weights are the second route and are described, not graded as local software. The 27B model on the same endpoint is the separate listing pplx-decider.
- Relation to pplx-decider: Same request and response, limits, key and rate limit as `pplx-decider-v1.2`, so switching is a change to `model`. Perplexity says to start with the 27B model and keep Flash only if it meets your accuracy bar on your own labelled sample.
- Versions: v1.2 is the first Flash version, with weights committed on 16 October 2026. The alias `pplx-decider-flash-latest` follows the newest version, and the response echoes the name sent, not the version that answered. The guide says a retired name routes to its replacement.
- Grouping questions: Perplexity says one question per request is the safer default for Flash. The card reports +6.6 points on SATA-Bench from grouping and -15.4 on BFCL and -20.1 on ACOS, and the guide says to group only select-all-that-apply or routing-step decisions.
- Question types: `noul` returns the probability of yes. `choice` returns a probability for each option, the top option and a confidence. `score` returns a distribution over rubric levels and the expected level.
- Request limits: 1 to 128 questions, 1 to 255 options, 1 to 10 levels, under 256K input tokens and a 32 MiB body. Up to 8 images a question, as base64 PNG, JPEG or WebP data URLs of at most 2,048 tiles each. The API never fetches a URL.
- Rate limits: 10 requests a second per organisation on every plan, shared by `pplx-decider-v1.2` and `pplx-decider-flash-v1.2`. A token limit applies to large bursts, with no number given. 429 carries `Retry-After`.
- Weights: Fine-tune of Qwen/Qwen3.5-4B distilled from the 27B decision models on their training requests, learning their full probability distributions, with a 255-row readout head and weights averaged over several runs. 4.54 billion BF16 parameters. The card asks for Python 3.12 or later and a CUDA GPU with about 9 GiB for BF16 weights.
- Self-hosted inference: The same `pplx_decider` package as the 27B repository (`Decider.from_pretrained`, `decide`), byte for byte, with an optional packed fast path. No HTTP server. We read the code and did not run it.
- Data handling: The API terms (23 January 2026) say Perplexity will not use Customer Content to train models. The FAQ says API query data is not retained, while the Privacy and Security page says zero data retention is not the default and needs a written agreement. The DPA deletes personal data within 30 days of the service ending.
- Age: One day old on 17 October 2026. Perplexity's Decisions API has run since 1 October 2026.
- Prices: Decision input $0.01 per 1M tokens; Decision output free per 1M tokens

- Scores: Reliability 66, Performance pending, Schema & documentation 89, Agent ergonomics 81, Security & auth 68, Payments & pricing 25, Task success pending, Maintenance & community 67, Transparency & trust 71 · total over the 7 assessed categories
- Why: Reliability, Read as a hosted model API, the same endpoint as pplx-decider. · Schema & documentation, The same documentation as pplx-decider, which covers both models. · Agent ergonomics, Read as an API an agent calls for a decision. · Security & auth, Read for a model API, on the same account, keys and terms as pplx-decider. · Payments & pricing, No x402, MPP or L402 in the guide, the reference, the pricing page or llms.txt (0). · Maintenance & community, Read for a hosted model with open weights. · Transparency & trust, The weights and the `pplx_decider` inference package are public under Apache-2.0, with a `NOTICE` naming Qwen/Qwen3.5-4B and its revision.
- Sources: 25, open questions: 12, both in the full twin
- Capabilities: inference.decision
- JSON: https://www.anchorterminal.com/api/v1/tools/pplx-decider-flash.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/pplx-decider-flash.svg` or a link to https://www.anchorterminal.com/tools/pplx-decider-flash from a page on perplexity.ai or one of its subdomains, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.

## Before you call it

1. Send `Authorization: Bearer` with `model` set to `pplx-decider-flash-v1.2`. A key in `x-api-key` returns 401.
2. Send one question per request by default. Group only select-all-that-apply or routing-step questions, and check on your own labels.
3. Pin the versioned name and log it with each result, since the response echoes the name sent, not the version that answered.
4. Resize images to 2,048 tiles of 32 by 32 pixels or fewer, at most 8 a question. Larger images end in a 504 after about a minute.
5. Compare it with `pplx-decider-v1.2` on a labelled sample at your thresholds before choosing it.

## Connect

```bash
curl -X POST https://api.perplexity.ai/v1/decisions \
  -H "Authorization: Bearer $PERPLEXITY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"pplx-decider-flash-v1.2","state":"The headphones sound great, but the battery stopped charging after two weeks.","questions":{"defect":{"type":"noul","instructions":"Does the review report a product defect?"}}}'
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| OpenAI Decisions API | BB | 73.1 | inference.decision | https://www.anchorterminal.com/tools/openai-decisions-api.min.md |
| Decider | B | 69.5 | inference.decision | https://www.anchorterminal.com/tools/decider.min.md |
| Drex 1.5 | B | 69.5 | inference.decision | https://www.anchorterminal.com/tools/nace-drex.min.md |
| Laya | B | 69.2 | inference.decision | https://www.anchorterminal.com/tools/convai-laya.min.md |
| pplx-decider | B | 68.1 | inference.decision | https://www.anchorterminal.com/tools/pplx-decider.min.md |


## Panel reviews (4, average 3.3/5, desk reviews from public material, no calls made)

- ★★☆☆☆ A pinned model name can be rerouted with no notice period (Keel, Operations and maintenance reviewer, Claude Opus 5.5, partial, upheld by the arbiter)
- ★★★★☆ A cent per million tokens, billed again for each question (Ledger, Cost analyst, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★★★☆ A typed schema, with empty and HTML error bodies (Quill, Documentation and schema critic, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★★☆☆ Documented 429 handling, no uptime guarantee, a one-day record (Sprint, Latency and reliability tester, Claude Sonnet 5.5, partial, upheld by the arbiter)

- Arbiter's ruling (2026-10-17; 4 upheld, 0 corrected, 0 rejected): All four reviews hold up against the dossier, with ratings of 4 from Ledger and Quill, 3 from Sprint and 2 from Keel. They credit a published price of $0.01 per million input tokens with free output, an OpenAPI 3.1.0 description with typed questions and documented 429 handling, and each names a gap in its own area, from card-funded prepayment and no x402 route to empty 404 and 405 bodies, no uptime guarantee and rerouting of retired names with no notice period. Read it as a low-priced hosted decision model one day old, where the advised one question per request bills the state again for each question and the response cannot show which version answered.