LLM Cost (OpenRouter)

pinakea uses OpenRouter to access cloud AI models. This page explains where OpenRouter credit can be used and how to keep spending controlled.

---

Where OpenRouter Credit Is Used

In Online Mode, OpenRouter is used for:

• Embeddings for semantic search
• Automatic item summaries
• AI titles
• AI tags
• Chat
• Full item summaries
• Day/daypart summaries
• AI tag consolidation

In Mixed Mode, OpenRouter is still used for:

• Chat
• Full item summaries
• Day/daypart summaries
• AI tag consolidation

In Mixed Mode, local processing handles embeddings, automatic item summary generation, AI titles, and AI tags.

Chat, full item summary generation, day/daypart summary generation, and AI tag consolidation still use OpenRouter.

Every online AI request is routed through OpenRouter Zero Data Retention providers. Chat uses the selected model’s fixed reasoning profile; summaries, titles, tags, tag consolidation, and embeddings do not use chat reasoning. Any text request that carries a reasoning setting requires an OpenRouter provider that supports that parameter.

Zero Data Retention on embeddings is not an embedding migration. Online Mode still uses qwen/qwen3-embedding-8b, so existing Online embeddings stay in the same vector space and do not need to be regenerated just because of this routing policy.

pinakea defaults to DeepSeek V4 Flash for online text generation: titles, summaries, tags, chat, and AI tag consolidation. You can override the current Set in Settings -> AI -> Online Model. New Sets and onboarding still default to DeepSeek. pinakea uses Qwen3 Embedding 8B for embeddings.

Model Price Reference

OpenRouter ZDR-listed base prices as of 1 July 2026:

Model	Context	Input $/1M	Output $/1M	Relative to DeepSeek
DeepSeek V4 Flash	~1M	$0.098	$0.196	default, lowest cost
Gemini 3.1 Flash Lite	~1M	$0.25	$1.50	~2.6x input / ~7.7x output
Grok 4.3	1M	$1.25	$2.50	~12.8x input / output
GPT-5.4	~1M	$2.50	$15.00	~25.5x input / ~76.5x output

DeepSeek is nearly orders of magnitude less expensive than the most expensive selectable model: GPT-5.4 output tokens are about 76.5x the DeepSeek output price at these base rates. Prices are not guaranteed. Long-context or tiered billing can be higher; OpenRouter notes higher tier pricing for Grok 4.3 above 200k total tokens, and provider-specific long-context pricing may apply for GPT-5.4.

Initial Library Processing

When you first add sources to pinakea in Online Mode, every item needs embeddings plus summaries, titles, and tags. This is usually the largest one-time OpenRouter usage event for a Set.

Actual usage depends on content length, provider pricing, and how much work is regenerated. Larger Sets naturally require more processing than smaller Sets. OpenRouter reports the real per-request cost in usage payloads, and pinakea uses that value when available.

Ongoing Usage

After initial processing, OpenRouter usage can come from:

• New items arriving through sources
• Chat conversations
• Search query embeddings in Online Mode
• Mode switches that regenerate AI content for a populated Set

Mode Switches

Switching modes can use OpenRouter credit.

• Online -> Mixed: Online embeddings generated with OpenRouter credit are replaced and cannot be reused for Mixed Mode. Existing automatic item summary records, titles, and tags are kept.
• Mixed -> Online: local embeddings, AI-generated summaries, AI-generated titles, and LLM tags are cleared and regenerated through OpenRouter credit.

Cost Control

You’re never at risk of surprise bills if you set limits:

1. Set a spending limit on your API key in OpenRouter’s dashboard.
2. Use a dedicated API key only for pinakea, so OpenRouter key usage matches pinakea usage.
3. Monitor spending in pinakea’s status bar and OpenRouter’s dashboard.
4. Avoid automatic top-up unless you explicitly want it.

See OpenRouter and BYOK for setup instructions.