Models inference pricing
Same models.
Cheaper hours.
Flexible work has a timing advantage. Move batch OCR, embeddings or evals to quieter hours and use lower dynamic rates.
Compare From prices below: the lowest price the dynamic multiplier can reach for each model. Prices move up or down within each model’s range.
€5 in free credits for every new customer.
Explore model ratesCompare model
inference rates.
From prices in EUR, updated 22 October 2026.
Prices move within a per-model range.
No models match your search.
Hover over or focus a model row to reveal Max; tap a price on mobile. Rounded console maxima, recorded on 21 October 2026, are indicative rather than guaranteed caps. — means unavailable.
Chat, code & reasoning
| Model | Input / 1M tokens | Output / 1M tokens |
|---|---|---|
 qwen3.6-35b-a3bReasoning with a long context · 256k contextAWQ-INT4 · 131 tok/s · 419 ms TTFT | From €0.0808Max €0.17 | From €0.542Max €1.15 |
qwen3.8-27bDense reasoning model · 64k contextW4A16 (INT4) · 58 tok/s · 328 ms TTFT | From €0.0816Max €0.21 | From €1.09Max €2.76 |
 gemma-4-26b-a4bLong-context text generation · 256k contextAWQ-INT4 · 140 tok/s · 370 ms TTFT | From €0.0398Max €0.09 | From €0.149Max €0.35 |
laguna-xs-2.1Coding and reasoning · 128k contextINT4 · 140 tok/s · 396 ms TTFT | From €0.0409Max €0.05 | From €0.0827Max €0.10 |
gpt-oss-20bReasoning · 128k contextMXFP4 · 180 tok/s · 372 ms TTFT | From €0.0183Max €0.05 | From €0.0697Max €0.17 |
hypernova-60bReasoning · 128k contextMXFP4 · 174 tok/s · 436 ms TTFT | From €0.043Max €0.12 | From €0.215Max €0.58 |
nemotron3-nano-30bLong-context reasoning · 256k contextAWQ-INT4 · 215 tok/s · 390 ms TTFT | From €0.034Max €0.05 | From €0.136Max €0.20 |
Vision & OCR
| Model | Input / 1M tokens | Output / 1M tokens |
|---|---|---|
qwen3-vl-30b-a3bImage understanding · 32k contextAWQ-INT4 · 179 tok/s · 515 ms TTFT | From €0.101Max €0.13 | From €0.402Max €0.50 |
nemotron3-omni-30bVision and audio understanding · 128k contextW4A16 (INT4) · 187 tok/s · 397 ms TTFT | From €0.06Max €0.16 | From €0.18Max €0.49 |
 deepseek-ocr-2Document OCR · 8k contextBF16 · 144 tok/s · 900 ms TTFT | From €0.0232Max — | From €0.0232Max — |
Text embeddings
| Model | Input / 1M tokens |
|---|---|
qwen3-embedding-8bText embeddings · 7k contextBF16 | From €0.00774Max <€0.01 |
Speech & audio
| Model | Per audio minute |
|---|---|
 voxtral-mini-3bTranscription and audio understanding · 8k contextBF16 · 25 tok/s (chat) · 855 ms TTFT | From €0.000774Max €0.0008 |
whisper-large-v3-turboSpeech to text · 30 s / segmentFP16 · 21.8× real-time (1 stream) | From €0.000155Max €0.0002 |
kokoro-82mText to speechFP32 · ~1× real-time (serial processing) | From €0.00399Max €0.0052 |
kyutai-tts-1.6b-en-frKyutai TTSBF16 | From €0.00543Max — |
Image generation
| Model | Per image |
|---|---|
flux-2-kleinImage generationBF16 · 4.15 s / image | From €0.00132Max €0.0036 |
These are catalogue rates for the listed models, not a tariff for every API route. For OCR and audio, check the route and billing unit before estimating your workload.
From prices in EUR, updated on 30 September 2026: the lowest price the dynamic multiplier can reach for each model. Confirm current rates and access in the console. Performance figures are indicative and shown per stream; TTFT means time to first token. Reasoning tokens count as output.
A rate you can
trace to the request.
For eligible models, the multiplier is set when you submit a request. Check recorded usage and charges in your console.
- 01
Start from the From price.
Each model has a From price for its billing unit.
- 02
Follow the hour’s rate.
Eligible models follow their GPU class’s hourly schedule, never below the From price.
- 03
Check your usage.
Review the request usage and charges recorded for your account.
A few useful
details.
Can dynamic prices go above list?
Yes. Prices follow demand and can rise above the list rate. Use the published GPU-class schedule when timing flexible work.
Does every model use dynamic pricing?
No. Dynamic pricing applies to eligible models. Other models use a fixed rate.
How is usage measured?
Start with the model and its API route. The catalogue uses tokens, audio minutes or generated images; dedicated parsing routes document per-image billing. Confirm the unit for your selected route in the console. Read the billing-unit guide.
What about AI Fabric pricing?
AI Fabric is opening gradually. GPU classes, scaling settings and billing terms are confirmed during onboarding. Join the waitlist with your workload; Models rates are not an AI Fabric GPU quote.
Explore AI Fabric