Pay per token. No subscriptions. No minimum spend.
Open-weight models. No content filtering. 85% to providers.
| Model | Category | Tier | Price / 1M tokens | Provider Earns (85%) |
|---|---|---|---|---|
| Qwen 2.5 0.5B | LLM | P2P | $1.00 (₹93) | $0.85 (₹79) |
| Llama 3.2 1B | LLM | P2P | $1.50 (₹139) | $1.28 (₹119) |
| Phi-3 Mini 4K | LLM | P2P | $2.50 (₹232) | $2.13 (₹198) |
Prices are per 1 million tokens. All models support text and code generation.
Payments processed in INR via Razorpay. INR amounts shown at live rate (₹92.81/$1 USD). Minimum charge: $0.001 per request.
Audio models (STT, TTS, VAD) coming soon.
Orchestrated workflows (research, code generation, agent, map-reduce) are billed at the same base model rate per step — no surcharge. Orchestration is a core differentiator, not an add-on.
1x
No surcharge per step
Budget cap
Set max_cost_usd per workflow
Free
Conversation memory
Use POST /v1/workflows/estimate to preview costs before running a workflow. Conversation memory is included at no extra charge.
Storage Limits (free tier)
100 chats
conversations per user
90-day
auto-cleanup of inactive data
Nataris runs open-weight models on real devices owned by real people. No moderation layer between your prompt and the model. No category restrictions on what you can build.
85% of what you spend goes directly to the person whose phone ran your job. The rest keeps the platform running.
No subscriptions, no minimum spend. Pay only for the tokens you use.
Provider devices only process what they're assigned — they never see your identity, history, or other requests. Your data is never used to train any AI model.
85% of every dollar goes directly to real people running inference on their devices, not to data centers.
$5 free credit on signup. No credit card required to start. Test all models before committing.