Torki API Models and Pricing

Overview

The Torki API has four models, the same four modes as the Torki app. You pay per token from prepaid credit, at a fixed price per model, and nothing else is charged.

The four models

Model	Best for	Thinking	Max output	Input / 1M tokens	Output / 1M tokens
torki-pulse	Fast, light work: extraction, classification, short answers	Off	32,768	$0.30	$1.20
torki-orbit	The everyday all-rounder: chat, writing, assistants	Off	49,152	$0.45	$1.80
torki-stellar	Deep reasoning: analysis, code, multi-step problems	On	73,728	$0.90	$3.60
torki-cosmos	Maximum depth: research, planning, the hardest problems	On	102,400	$1.20	$4.80

• Context window: 131,072 tokens on all four, prompt plus output. A longer prompt is refused with 400 context_length_exceeded before it reaches a model.
• Inputs: all four read text, images and video. See Images, Video and Object Detection.
• Default max_tokens: 8,192 on Pulse, 16,384 on Orbit, and the whole budget on Stellar and Cosmos.
• Default temperature when you don't send one: 0.3 on Pulse, 0.6 on Orbit, 0.7 on Stellar and Cosmos.

GET /v1/models lists the models with their context window, max output, thinking, input types and prices. The Models page in Torki Platform shows the same.

Which one to choose

Start with torki-pulse for high-volume or latency-sensitive work and torki-orbit for general chat and writing. Use torki-stellar or torki-cosmos when the task needs reasoning: they think before answering, which takes longer and produces more output tokens. See Thinking Models: Stellar and Cosmos.

What you are charged

• Input tokens — the whole prompt, including images and video frames, at the model's input price.
• Output tokens — the answer, plus the thinking on Stellar and Cosmos. Thinking tokens bill as output. usage.completion_tokens already includes them (also shown as usage.completion_tokens_details.reasoning_tokens when the server reports it), so they are never charged twice.
• Each request is charged after it finishes, from its token counts, rounded up to the nearest $0.000001.
• If a stream is interrupted, the tokens produced up to that point are charged.
• A request that fails before any output is not charged. Neither is a request refused by content moderation.
• If Stellar or Cosmos thinks through its whole budget without answering, the API answers once more without thinking. The failed attempt is not charged, up to 5 times a day per organization; after that it bills like normal tokens.
• Torki adds a short line of its own ahead of your system messages. Its tokens are not counted in usage.prompt_tokens and are never billed.
• Promotional credit from Torki is spent before credit you bought.

Worked examples

• Text: a 2,000-token prompt and a 500-token answer on torki-orbit cost 2,000 × $0.45 + 500 × $1.80 per million tokens = $0.0009 + $0.0009 = $0.0018.
• An image: about one input token per 32 × 32 pixels. A 640 × 480 photo is about 300 tokens. A phone photo at the default detail is about 1,750 tokens: about $0.0005 on Pulse, or $0.0021 on Cosmos.
• A video: 8 frames of a 16:9 video are about 2,700 input tokens: about $0.0008 on Pulse or $0.0012 on Orbit.

Your own figures are on the Usage and Cost pages, and each request's cost is in Logs. See Usage, Cost and Logs in Torki Platform.

FAQ

Are prices in US dollars?

Yes. Token prices and your credit balance are always in US dollars. If your billing address is in India you buy credit in rupees, but the credit you get is the pack's dollar value.

Can I turn thinking off on Stellar or Cosmos?

Yes. Send "thinking": false for faster, cheaper answers. See Thinking Models: Stellar and Cosmos.

Is there a Batch API discount?

The Batch API, at 50% of these prices, is coming soon. It is not open yet.