Token Pricing
Transparent pricing for all AI models and services across input and output tokens.
Transparent pricing for all AI models and services across input and output tokens.
Language models
Chat completion and text generation models.
| Model | Description | Input / 1M | Output / 1M |
|---|---|---|---|
| GreenL | Lightweight, efficient language model. | €0.25 | €0.80 |
| GreenR | Reasoning-optimized language model. | €0.35 | €0.95 |
| gemma4 | Long-context multimodal model (256k). Public Preview. | €0.50 | €1.50 |
Chat models
Direct provider chat models available through the GreenPT API proxy, per 1M tokens.
The Cached input column is what a prompt-cache hit costs on that model. Repeated prompt prefixes are served from cache automatically and billed at that lower rate; a dash means the model has no cache discount and bills every input token at the full rate. See Prompt caching. Output tokens are never discounted by prompt caching.
| Model | Input / 1M | Cached input / 1M | Output / 1M |
|---|---|---|---|
glm-5.2 | €1.10 | €0.275 | €4.40 |
glm-5.2-caveman / -honey / -ponytail (compression variants, incl. -lite and -ultra) | €1.10 | €0.275 | €4.40 |
kimi-k3 | €3.30 | €0.825 | €16.50 |
kimi-k2.7-code | €0.79 | €0.165 | €3.85 |
kimi-k2.6 | €0.66 | €0.22 | €3.75 |
minimax-m2.5 | €0.17 | €0.055 | €0.99 |
deepseek-v4-flash-0731 | €0.14 | €0.04 | €0.35 |
llama-3.3-70b-instruct | €1.10 | – | €1.10 |
gemma-3-27b-it | €0.30 | – | €0.60 |
devstral-2-123b-instruct-2512 | €0.50 | – | €2.40 |
qwen3-235b-a22b-instruct-2507 | €0.90 | – | €2.70 |
mistral-small-3.2-24b-instruct-2506 | €0.20 | – | €0.40 |
qwen3-coder-30b-a3b-instruct | €0.25 | – | €0.95 |
qwen3.5-397b-a17b | €0.70 | – | €4.35 |
qwen3.6-35b-a3b | €0.30 | – | €1.80 |
mistral-medium-3.5-128b | €1.80 | – | €9.00 |
holo2-30b-a3b | €0.35 | – | €0.85 |
pixtral-12b-2409 | €0.25 | – | €0.25 |
gpt-oss-120b | €0.20 | – | €0.70 |
voxtral-small-24b-2507 | €0.20 | – | €0.45 |
Vector & search models
Embedding and reranking services.
Note: embedding and reranking models only charge for input tokens. There are no output token costs for these services.
| Model | Description | Price / 1M tokens |
|---|---|---|
| Green Embedding | Text vectorization and semantic search. | €0.20 |
qwen3-embedding-8b | Larger multilingual embedding model. | €0.25 |
| Green Rerank | Document reranking and relevance scoring. | €0.12 |
Speech-to-text models
Audio transcription services (priced per second).
New pricing, now live: the per-hour model rates below are updated and in effect today. The add-on features are a separate launch promo and free for now.
Limited promo: 50% off (July & August 2026)
All speech-to-text model rates are half price through 31 August 2026. The promo column below shows the discounted rate.
| Model | Mode | Regular rate | Promo (Jul–Aug 2026, −50%) |
|---|---|---|---|
| GreenS | Pre-recorded audio (batch) | €0.23 / hour | €0.12 / hour |
| GreenS | Live audio (real-time) | €0.31 / hour | €0.16 / hour |
| GreenS Pro | Monolingual: pre-recorded | €0.23 / hour | €0.12 / hour |
| GreenS Pro | Monolingual: live audio | €0.31 / hour | €0.16 / hour |
| GreenS Pro | Multilingual: pre-recorded | €0.28 / hour | €0.14 / hour |
| GreenS Pro | Multilingual: live audio | €0.37 / hour | €0.19 / hour |
Additional features
Launch promo: the speech-to-text add-ons below are free right now. The prices shown are the standard per-hour rates that apply once the promo ends; you are not charged for them today.
| Feature | Rate (promo: free now) |
|---|---|
| Redaction | €0.10 / hour |
| Entity detection | €0.08 / hour |
| Streaming diarization (live audio) | €0.10 / hour |
| Keyterm prompting | €0.07 / hour |
Speaker diarization is included free for pre-recorded audio. Multichannel audio is billed per channel, so 2-channel audio is charged at double the per-hour rate.
Note: embedding, reranking, and speech-to-text services don't generate output tokens. They're billed only on input.
Token usage examples
A request is billed on the tokens it sends plus the tokens it generates, so the cost of a call follows from the rates above. As a rule of thumb, 1,000 tokens is roughly 750 words of English text. A few worked examples:
| Example | Model | Input tokens | Output tokens | Cost |
|---|---|---|---|---|
| A short question and answer | green-l | 300 | 150 | €0.0002 |
| A page of text summarized | gemma4 | 8,000 | 500 | €0.0048 |
| An analysis with reasoning | green-r | 2,000 | 800 | €0.0015 |
| A coding turn on a warm cache | glm-5.2 | 20,000 (15,000 cached) | 1,200 | €0.0149 |
| A full repository file read into the prompt | glm-5.2 | 100,000 | 2,000 | €0.1188 |
Two details the examples above show:
- Reasoning tokens are output tokens. What the model generates while
thinking is included in
completion_tokensand billed at the output rate. - Cached input is cheaper. In the coding example, 15,000 of the 20,000 input
tokens are a repeated prompt prefix, billed at
glm-5.2's cached rate of €0.275 instead of €1.10. See Prompt caching.
Services that don't generate tokens are billed the same way, on input alone: embedding 500,000 tokens with Green Embedding costs €0.10, reranking 200,000 tokens with Green Rerank costs €0.024, and transcribing an hour of pre-recorded audio with GreenS costs €0.23 (€0.12 during the promo).
Agents
Agent runs have no per-run price; a run is billed as the sum of what it consumes, at the same rates as direct API calls:
| Component | Billed as |
|---|---|
| Reasoning steps and image analysis | Input + output tokens of the model used |
| Meeting transcription | Speech-to-text seconds |
web_search, web_crawl, document_parse | Free |
Creating a run requires a positive credit balance; see Agents → Billing for how costs accrue during a run.
Pricing notes
- All prices are in EUR and exclude applicable taxes.
- Token counting follows OpenAI-compatible standards.
- On models with a cached-input rate, repeated prompt prefixes are billed at
that rate instead of the full input rate. It applies automatically, and the
response reports the count on
usage.prompt_tokens_details.cached_tokens. See Prompt caching. - Speech-to-text pricing is based on audio duration, not tokens.
- Output tokens are typically priced higher than input tokens.
- Embedding and reranking models only charge for input tokens (no output costs).
- Volume discounts available for enterprise customers.
- Prices subject to change with 30 days notice.