AI Models & Cost Levels

Model cost levels

Each level has a rate expressed in HC per million weighted tokens. Level 0 is free — models run on our own GPU infrastructure. Levels 1-5 are premium models, priced according to their computational cost.

Weighted tokens explained

When you send a request, the platform doesn't count raw tokens. Instead it calculates weighted tokens to reflect the real compute cost of each token type:

Worked example

That's 6 HC for a typical request. With a Starter plan (1,500 HC/month) you could make roughly 250 similar requests. With a Plus plan (6,000 HC/month) that's ~1,000 requests.

Economy mode

When you run out of HC — whether you're on the free Trial or you've exhausted your monthly allowance — the platform automatically switches to economy mode. In this mode, only Level 0 (Free) models are available, at no cost. You keep working without interruption, just with the free model catalogue.

Level 0 includes Qwen 3.6 35B, running on our own GPU infrastructure. Enough for most everyday tasks.

Bring your own API keys

You can connect your own API keys for any compatible provider or any OpenAI-compatible endpoint. When you use your own keys, requests don't consume any HC — you pay the provider directly at their rates. This is ideal if you already have credits with a provider or want to use models not in our catalogue.

Learn more in How Credits Work.

Complete model catalogue

All models available on Hermes Haven, organised by cost level. HC rates shown per million weighted tokens. Non-LLM models (TTS, STT, image, video, embeddings) show per-unit pricing — see dashboard for details.

FAQ

What are weighted tokens?

Weighted tokens are a measure of actual compute cost. Input tokens count ×1, cached input ×0.1, output tokens ×5, and reasoning tokens ×5. This reflects the fact that generating text is more expensive than reading it, and that cached context is nearly free to reuse.

How much does a typical request cost?

A request with 1,000 input tokens and 500 output tokens using GLM-5.2 (Level 3) produces 3,500 weighted tokens, which costs 6 HC. With a Starter plan (1,500 HC/month), that's roughly 250 similar requests per month.

What happens when I run out of HC?

The platform switches to economy mode. Only Level 0 (Free) models are available, at no cost. You can keep working with the free model until your HC resets at the next billing cycle or you top up.

Are Level 0 models really free?

Yes. Level 0 models cost 0 HC per million weighted tokens. They run on our own GPU hardware and subsidised cloud capacity. They're available on every plan, including the free Trial, and remain available in economy mode when all credits are exhausted. The free model is Qwen 3.6 35B.

Can I use my own API keys?

Yes. You can bring your own API keys for any compatible provider or any OpenAI-compatible endpoint. Requests made with your own keys don't consume any HC — you pay the provider directly. This lets you use models outside our catalogue at provider rates.

How are HC rates calculated?

HC rates are set based on the computational cost of each model. For LLM models with separate input/output pricing, we use the weighted token formula: blended = (input + 5 × output) / 6, reflecting that output is 5× more expensive than input.

How do HC top-ups work?

You can buy HC top-up bundles at any time: €10 → 10,000 HC, €50 → 50,000 HC (10% bonus), €100 → 100,000 HC (15% bonus). Top-ups never expire. See Pricing for details.

What's the minimum charge per request?

The minimum charge is 1 HC. If a request's calculated cost is below 1 HC (e.g. a very short Level 1 request), it's rounded up to 1 HC. Level 0 requests always cost 0 HC regardless of size.

Why do some models show per-unit pricing instead of HC/M?

TTS (text-to-speech), STT (speech-to-text), image generation, video generation, music, and voice design models are priced per unit (per character, per minute, per image, per second, per request) rather than per million tokens. See dashboard for per-unit pricing details.