Model APIs

Every route, and what it costs.

The whole P/9 catalogue, across both serving pools, priced per million tokens in rupees. One endpoint, one key, one bill.

Read the docs

Global pass-through

Relayed by the gateway to partner capacity outside the PK NPU pool. Same endpoint, same keys, billed in rupees like every other route.

PK Endpoints

Served from P/9's own Ascend 910B NPU capacity in Pakistan, the platform's native hardware. Open with the same keys and the same rupee billing as every other route.

gpt-oss 120BLive
OpenAI · PK NPU · Chat completions
Input, Rs / 1M
Rs 30.08
Cached input
Rs 30.08
Output, Rs / 1M
Rs 150.39
Route id
paranine/gpt-oss-120b(high)
Qwen3.8 27BLive
Qwen team, Alibaba · PK NPU · Chat completions
Input, Rs / 1M
Rs 80.00
Cached input
Same as input
Output, Rs / 1M
Rs 650.00
Route id
paranine/qwen-3.8-27b
DeepSeek-OCR 2Live
DeepSeek · PK NPU · Chat completions
Document OCR: send a scan or photo of a page and get its text back as Markdown, with headings, tables and line items kept.
Input, Rs / 1M
Rs 9.02
Cached input
Same as input
Output, Rs / 1M
Rs 9.02
Route id
paranine/DeepSeek-OCR2
Laya DecisionLive
ConvAI Innovations · PK NPU · Decisions
Ask yes/no, multiple-choice and rating questions about a piece of text and get an answer with a probability for each option.
Input, Rs / 1M
Rs 9.94
Cached input
Same as input
Output, Rs / 1M
No output tokens
Route id
paranine/laya-decision
Qwen3 Embedding 8BLive
Qwen team, Alibaba · PK NPU · Embeddings
Input, Rs / 1M
Rs 2.78
Cached input
Same as input
Output, Rs / 1M
No output tokens
Route id
paranine/qwen3-embedding

Every model route P/9 serves, grouped by serving pool, with the identifier to put in a request’s model field and the price per million tokens in Pakistani rupees.
Model	Route id (the model string)	Input, Rs / 1M	Cached input, Rs / 1M	Output, Rs / 1M
Global pass-through Relayed by the gateway to partner capacity outside the PK NPU pool. Same endpoint, same keys, billed in rupees like every other route.
gpt-oss 120BLive OpenAI · Global · Chat completions	`paranine/gpt-oss-120b(Global)`	Rs 30.08	Rs 30.08	Rs 150.39
Kimi K3Live Moonshot AI · Global · Chat completions	`paranine/Kimi-K3(Global)`	Rs 902.34	Rs 90.23	Rs 4,511.70
GLM-5.3Live Z.ai · Global · Chat completions Z.ai's newest model for agentic engineering: coding, tool use and long-horizon agent work.	`paranine/GLM-5.3(Global)`	Rs 421.09	Rs 42.11	Rs 1,323.43
GLM-5.3 FastLive Z.ai · Global · Chat completions GLM-5.3 tuned for speed: the same model on a faster serving tier. A preview at the provider, rate-limited across all of Paranine.	`paranine/GLM-5.3-Fast(Global)`	Rs 631.64	Rs 63.16	Rs 1,985.15
DeepSeek V4.1 FlashLive DeepSeek AI · Global · Chat completions	`paranine/DeepSeek-V4.1-Flash(Global)`	Rs 90.23	Rs 2.11	Rs 360.94
Kimi K2.7 CodeComing soon Moonshot AI · Global · Chat completions	`paranine/Kimi-K2.7-Code(Global)` Proposed, not registered yet	Rs 285.74	Same as input	Rs 1,203.12
DeepSeek-V4-Flash-0731Coming soon DeepSeek AI · Global · Chat completions An open-weight mixture of experts, 284B total and 13B active, with a one-million-token context and selectable reasoning effort.	`paranine/DeepSeek-V4-Flash-0731(Global)` Proposed, not registered yet	Rs 39.10	Same as input	Rs 78.20
DeepSeek V4 Pro 0813Coming soon DeepSeek AI · Global · Chat completions The 0813 release of DeepSeek's 1.6T-parameter open frontier model.	`paranine/DeepSeek-V4-Pro-0813(Global)` Proposed, not registered yet	Rs 397.03	Same as input	Rs 1,191.09
DeepSeek V4 ProComing soon DeepSeek AI · Global · Chat completions	`paranine/DeepSeek-V4-Pro(Global)` Proposed, not registered yet	Rs 523.36	Same as input	Rs 1,046.72
Nvidia Nemotron 3 UltraComing soon Nvidia · Global · Chat completions	`paranine/Nemotron-3-Ultra(Global)` Proposed, not registered yet	Rs 180.47	Same as input	Rs 721.87
InklingComing soon Thinking Machines Lab · Global · Chat completions A multimodal mixture of experts, 975B total and 41B active, reasoning over text, images and audio at a 256k context.	`paranine/Inkling(Global)` Proposed, not registered yet	Rs 300.78	Same as input	Rs 1,218.16
Inkling-SmallComing soon Thinking Machines Lab · Global · Chat completions The smaller, open-weight Inkling: faster and cheaper to run, with the same native text, image and audio input.	`paranine/Inkling-Small(Global)` Proposed, not registered yet	Rs 150.39	Same as input	Rs 360.94
PK Endpoints Served from P/9's own Ascend 910B NPU capacity in Pakistan, the platform's native hardware. Open with the same keys and the same rupee billing as every other route.
gpt-oss 120BLive OpenAI · PK NPU · Chat completions	`paranine/gpt-oss-120b(high)`	Rs 30.08	Rs 30.08	Rs 150.39
Qwen3.8 27BLive Qwen team, Alibaba · PK NPU · Chat completions	`paranine/qwen-3.8-27b`	Rs 80.00	Same as input	Rs 650.00
DeepSeek-OCR 2Live DeepSeek · PK NPU · Chat completions Document OCR: send a scan or photo of a page and get its text back as Markdown, with headings, tables and line items kept.	`paranine/DeepSeek-OCR2`	Rs 9.02	Same as input	Rs 9.02
Laya DecisionLive ConvAI Innovations · PK NPU · Decisions Ask yes/no, multiple-choice and rating questions about a piece of text and get an answer with a probability for each option.	`paranine/laya-decision`	Rs 9.94	Same as input	No output tokens
Qwen3 Embedding 8BLive Qwen team, Alibaba · PK NPU · Embeddings	`paranine/qwen3-embedding`	Rs 2.78	Same as input	No output tokens

How to read this table.

Coming soon means the tag, not the price: A route listed ahead of opening answers 404 until the tag comes off. Its rate is published now so an integration can be costed before it ships, and it is the rate the meter will use on day one.
A proposed id is the shape, not the string: Some routes carry an id P/9 intends to publish but has not registered on the gateway yet. Those say so under the id. Write against them if you like, but expect to change the string once the route opens.
Prices are per million tokens, input and output metered apart: A request is billed on the tokens it actually consumes: the prompt at the input rate, the completion at the output rate. Embedding routes return vectors rather than tokens, so they have no output rate at all.
There is no listing endpoint yet: GET /v1/models is not served, so a stock OpenAI SDK's models.list() returns 404. This page, the dashboard's Model APIs panel and a key's allowed_models are the sources of truth until it ships.

Every route, and what it costs.

Global pass-through

PK Endpoints

Global pass-through

PK Endpoints

How to read this table.

Cost-effective inference for every leading open model.