moondream3.1 9b a2b
unofficialvision
moondream3.1 9b a2b specifications
| Context window | — |
|---|---|
| Max output tokens | 28672 |
| Release date | — |
| Input | Text, Image |
| Output | Text |
moondream3.1 9b a2b pricing in rubles
You pay for what you use, per token and per unit. The final request cost is returned in the API response.
| Type | Price |
|---|---|
| Input tokens | 60 ₽ per 1M tokens |
| Output tokens | 200 ₽ per 1M tokens |
Prices are in rubles and may change with exchange rates.
What moondream3.1 9b a2b can do
- Image understanding
- Reasoning mode
moondream3.1 9b a2b API: connection and code examples
The service exposes a single OpenAI-compatible API. Set our base_url and the key from your dashboard.
cURL
curl https://api.neuralbridge.ru/v1/vision \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "moondream/moondream3.1-9b-a2b-08130", "image": "data:image/png;base64,...", "prompt": "Что на картинке?"}'API endpoints
POST /v1/visionprimary | Vision Image understanding: captioning, questions about an image, object and point detection. |
Supported request parameters
| Parameter | Default | Allowed values | Description |
|---|---|---|---|
temperature | 1 | — | Controls how varied the answer is: higher values make the output more random, lower values more predictable. |
top_p | 1 | — | Nucleus sampling: only the most probable tokens whose probabilities add up to P are considered. |
max_tokens | — | — | Upper limit on the number of tokens the model may generate in its answer. |
presence_penalty | 0 | — | Penalizes tokens that already appeared in the text, nudging the model toward new topics. |
frequency_penalty | 0 | — | Penalizes tokens in proportion to how often they already appeared — fights repetition. |
stop | — | — | List of stop strings: generation halts as soon as the model produces any of them. |
seed | — | — | Pins randomness: repeating a request with the same seed and parameters returns the same result. |
response_format | — | — | Response format: plain text or strict JSON following a given schema. |
stream | false | — | Deliver the answer as a stream while it is generated instead of one chunk at the end. |