Yes. Available to paid accounts. Ask on pilot@fikraapi.co.ke if you want to try it before the catalogue opens.

Perplexity Decider v1.3 on Fikra API

Q: What is the difference between v1.3 and Flash v1.3?

Flash is a smaller distilled version for high-volume classification; v1.3 is for nuance-critical decisions.

The model

A specialist, not a generalist.

Decider reads text or images, takes a set of questions, returns a probability per option. Yes/No, multiple choice, and scored levels. Same API shape as any OpenAI-shaped endpoint on Fikra.

Where it fits

Routing: which LLM to hand this off to — tiny, medium, frontier.
Beam / quality scoring: grade candidate generations before a human or a stronger model reviews.
Agent gates: continue, stop, escalate, fetch more context.
Policy filters: classify content before it hits a paid API downstream.
Claim verification: is this claim supported by the document? Perplexity reports its own evaluation results; see their announcement. Not independently verified by Fikra.

Technical shape

Parameters

27B (v1.3), 4B (Flash v1.3). Built on Qwen3.8-27B and a distilled Qwen3.5-4B.

License

Apache 2.0. Use the open weights in your own stack if you want to run it there.

Context

1M+ tokens (same per-input limit as a normal chat model).

Price

$0.02 per 1M input (v1.3), $0.01 per 1M (Flash). Output free.

Availability

Available via Fikra API on request. Ask on [email protected] for pilot access.

Example

One input, many questions.

# Perplexity Decisions API; same key as every other model on Fikra curl -X POST https://api.fikraapi.co.ke/v1/decider \ -H "Authorization: Bearer fk_live_..." \ -H "Content-Type: application/json" \ -d '{ "input": {"text": "Customer claims a refund for the July invoice."}, "questions": { "urgent": ["yes", "no"], "sentiment": ["angry", "neutral", "happy"], "next_action": ["refund", "escalate", "dismiss"] } }'

Response shape: one probability per option per question. Threshold anything below your floor.

Ask for access

Decider is on the pilot roadmap. Send a note to [email protected] with the workload you want to test and we will route you.

Open weights, your stack. If self-hosting fits better, pull the weights from Hugging Face and serve them on your own GPUs. Fikra hosts the API for teams that do not want to.

Questions

Common questions.

Is Decider on Fikra?

Yes. Available to paid accounts. Ask on [email protected] if you want to try it before the catalogue opens.

What is the difference between v1.3 and Flash v1.3?

Flash is a smaller distilled model for high-volume classification. Use v1.3 for nuance-critical decisions.

What's the context window?

1M+ tokens. Same class of context as the mainstream chat models, but the decision request lives inside it.

Can I self-host?

Yes. Apache 2.0; weights on Hugging Face. You can run it entirely in your own infrastructure.