Pay for what you send. The scores are free.
One price per million input tokens. Input is the content plus your questions — no system prompt, no repeated context, no output charge. New accounts start with $4.50 of free credit (50M tokens).
A launch model, and a larger one on request
Both are Google DeepMind’s open-weights Gemma 4 models, optimized by spinf for scoring at scale. The 12B is the launch model; the 31B, for harder questions, is available on request: contact sales to enable it on your account.
How a request is billed
spinf 12B
spinf-12bThe launch model: Google DeepMind’s open-weights Gemma 4 12B, optimized by spinf for scoring at scale.
Model idspinf-12b or google/gemma-4-12B$0.09
/ 1M input tokens
spinf 31B
spinf-31bGemma 4 31B, optimized the same way, for harder questions and fewer examples. Available on request.
Model idspinf-31b or google/gemma-4-31BContact Sales
Ask for accessSet for the best results today
Limits are set to keep performance high during the first deployments. The system has been tested with content up to 64K tokens and longer questions, but in that configuration the current models become less effective — it is still under investigation.
| Content per input | 4,096 tokens | 8,192 under evaluation |
| Minimum per call | 1,000 tokens | all processed tokens of the call together |
| Questions per content | Up to 10,000 | templates × combinations; up to 2M processed tokens and 2 MB per request |
| Tokens per question | 256 max | the filled template |
| Answers per question | Up to 1,000 | 1–5 tokens each |
| Output | Free | p and empty floor per answer, residual per question |
| Concurrent requests | By credit balance | see the table below |
Concurrency by credit balance
One concurrent request per $100 of credit, at least one. Your current limit is shown in the console.
Real time, and beyond a trillion tokens
Reserved endpoints for real-time decisions, and for projects over 1T tokens dedicated deployments and file-based batch analysis, so we can match the speed requirements of even the largest projects. Priced per project.
Reserved endpoints for real time
The on-demand service is built for throughput. For decisions in live flows (ticket routing, moderation before publishing, agents) we reserve capacity for you: always warm, sized to your traffic, pinned with one request field.
File-based batch (S3 drop)
Drop files in a bucket, get scores back in batch. Built for projects over a trillion tokens, where throughput matters more than latency.
Dedicated deployments
Capacity sized to your deadline, so even the largest backfills finish on schedule.
On your premises
The engine runs on datacenter-grade and consumer hardware (24 GB GPU minimum). For financial institutions and government services that need an extra level of privacy and security.
Put the full power of the leading open-weights language models to work at hyperscale: we optimize them for your use cases, so analyses that used to be too slow or too expensive to run become routine.