Developers

Ship AI products from Pakistan.

One API, every leading open-source model, billed in rupees. New organisations start with Rs 1,000 of free credits. No credit card required at sign-up.

Read the docs(opens in a new tab)
Start in thirty seconds

Point any OpenAI SDK at P/9 and call a model.

Point any OpenAI-compatible SDK at sovereign.paranine.com/v1, swap in your P/9 API key, and call the model you need. Existing agents, tool chains and RAG pipelines work without code changes.

Set once, in your shell
export P9_BASE_URL=https://sovereign.paranine.com
export P9_API_KEY=your_key_here
curl "$P9_BASE_URL/v1/chat/completions" \
  -H "Authorization: Bearer $P9_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "paranine/gpt-oss-120b(Global)",
    "messages": [
      { "role": "user", "content": "Explain the P/9 router in one sentence." }
    ]
  }'

Powered by optimised open-source AI

One API. Every leading open-source AI model.

Every route P/9 serves today, priced per million tokens in rupees. Every card is the same catalogue the dashboard bills against.

Global

gpt-oss 120B

OpenAI · 117B total, 5.1B active

Chat completions · 131,072 tokens context

Input / 1M

Rs 30.08

Output / 1M

Rs 180.47

Tap to open the model card

Global

Kimi K3

Moonshot AI · 2.8T total, 104B active

Chat completions · 1,048,576 tokens context

Input / 1M

Rs 902.34

Output / 1M

Rs 4,601.93

Tap to open the model card

Global

GLM-5.3

Z.ai

Chat completions

Input / 1M

Rs 421.09

Output / 1M

Rs 1,365.54

Tap to open the model card

Global

GLM-5.3 Fast

Z.ai

Chat completions

Input / 1M

Rs 631.64

Output / 1M

Rs 2,048.31

Tap to open the model card

Global

DeepSeek V4.1 Flash

DeepSeek AI

Chat completions

Input / 1M

Rs 90.23

Output / 1M

Rs 363.04

Tap to open the model card

PK NPU

Qwen3.8 27B

Qwen team, Alibaba · 27B, dense

Chat completions · 262,144 tokens context

Input / 1M

Rs 80.00

Output / 1M

Rs 650.00

Tap to open the model card

PK NPU

DeepSeek-OCR 2

DeepSeek

Chat completions

Input / 1M

Rs 9.02

Output / 1M

Rs 9.02

Tap to open the model card

PK NPU

Laya Decision

ConvAI Innovations

Decisions

Input / 1M

Rs 9.94

Output / 1M

No output tokens

Tap to open the model card

PK NPU

Qwen3 Embedding 8B

Qwen team, Alibaba · 8B

Embeddings · 32,768 tokens context

Input / 1M

Rs 2.78

Output / 1M

No output tokens

Tap to open the model card

View all models
Models available today

Chat and reasoning.

Every route P/9 serves is priced per million tokens in PKR. Pick the model that fits the task; the gateway handles the rest.

Kimi K3 (Moonshot AI)

2.8T total / 104B active params. 1M-token context. Best-in-class on hard reasoning in maths, science, and code. Long-horizon agentic work.

GLM-5.3 and GLM-5.3 Fast (Z.ai)

Strong on coding, tool use, and sustained execution across long tasks. GLM-5.3 Fast is the same model on a faster serving tier, a rate-limited preview.

gpt-oss 120B (OpenAI)

117B total / 5.1B active. Apache 2.0. Agentic workflows, reasoning-heavy work across maths, science, and code.

DeepSeek V4.1 Flash (DeepSeek AI)

Fast, cost-effective general-purpose completions.
Specialised

Beyond chat and reasoning.

Full model catalogue with per-million-token pricing in PKR: paranine.com/models.

DeepSeek-OCR 2

Send a scan or photo, get structured Markdown back with headings, tables, and line items.

qwen3-embedding

Production-grade embeddings for RAG and search.

Laya Decision

Ask yes/no, multiple-choice and rating questions about a piece of text and get an answer with a probability for each option.

TTS, STT, and Image/OCR models

Served through the same unified API.
Intelligent model routing

The P/9 Router predicts and sends each request to the best-fit open-source model, cutting cost while holding quality.

Incoming request

Simple chat

P/9 Router
  • p9-fastQwen3-8B
  • p9-coderQwen3-Coder
  • p9-reasongpt-oss 20B
  • p9-largeDeepSeek V3.2
  • p9-embedbge-large

Per-request routing

A P/9 agent reads each prompt and task, then routes to the best-suited model in the fleet.

Quality preserved

Top-tier answers by sending hard requests to stronger models, easy ones to lighter, cheaper models.

Cost driven down

Most traffic is served by lightweight open-source models, so you only pay for heavyweight compute when it matters.

Router aliases

Hard requests go to heavyweight models. Simple ones go to lighter, cheaper models.

The P/9 Router reads each incoming prompt and sends it to the best-fit open-source model in the fleet. Quality stays high. Cost goes down.

p9-fast

Routes to Qwen3-8B for lightweight, high-speed completions.

p9-coder

Routes to Qwen3-Coder for code generation and review.

p9-reason

Routes to gpt-oss 20B for reasoning-heavy tasks.

p9-large

Routes to DeepSeek V3.2 for complex, context-heavy work.

p9-embed

Routes to bge-large for embedding and retrieval.
What the gateway gives you

P/9 is not just an inference endpoint. It is a sovereign gateway with per-key controls, cost attribution, and observability built in.

API keys per project

Scoped by allowed models, with expiry and a rupee credit limit. One key per project, one invoice for the org.

Cost tracking per request

Every response returns its exact rupee cost in debug metrics. No surprise bills.

Hard spend caps

Set a credit limit per key. At the cap, that key stops with a 402. Everything else keeps running.

Token attribution

Track usage per team, per application, per feature, per client. Built for freelancers billing by project and enterprises chargebacking across departments.

Model access control

Scope each key to specific models. Anything outside scope gets a 403.

Audit log

Append-only record of every admin action, exportable as CSV.
Pricing

Pay-as-you-go. Top up your prepaid balance in PKR. Spend it per token at published rates.

No subscription, no hidden platform fee, no exchange-rate exposure. Full pricing: paranine.com/pricing.

Prices quoted, billed, and settled in PKR

Same currency from top-up to spend.

Prepaid balance

Budget exhaustion fails closed. You never run on credit.

No API call limit

Rate limits are per-key and adjustable, not platform-wide.

Free credits

Rs 1,000 on sign-up to test any model in the catalogue.
Performance

The gateway is a Go service on the request path: authenticate, check scope and budget, meter tokens, relay.

Its overhead is benchmarked with an open, reproducible harness against a deterministic mock upstream.

2.28ms

P50 latency through the gateway

6.75ms

P99 latency through the gateway

4,520req/s

Sustained throughput on 2 vCPUs

3.4MB

Docker image size

Sustained requests per second, one instance of each gateway (higher is better)

  • P/94,520 req/s
  • GoModel3,961 req/s
  • Bifrost2,560 req/s
  • Portkey993 req/s
  • LiteLLM288 req/s

Benchmarked against LiteLLM (288 req/s), Portkey (993 req/s), Bifrost (2,560 req/s), and GoModel (3,961 req/s) on the same host and mock upstream. Harness is open source.

Benchmark repo: github.com/ENTERPILOT/ai-gateway-reproducible-benchmark

Get started

Get your API key at paranine.com.

Docs at docs.paranine.com. Questions: info@paranine.com.

Cost-effective inference for every leading open model.