Models & pricing
Every hosted app on SkillSafe runs on one of these models. Usage is billed per token in credits — $1 = 10,000 credits — at the rates below, so you always know what a run costs before you ship or use an app.
| Model | Provider | Input / 1M tokens | Output / 1M tokens | Max output | Time limit |
|---|---|---|---|---|---|
| Claude Haiku 4.5 claude-haiku-4-5 | Anthropic | $1.20 | $6.00 | 4,096 tokens | 60s |
| Claude Sonnet 5 claude-sonnet-5 | Anthropic | $3.60 | $18.00 | 8,192 tokens | 120s |
| Claude Sonnet 4.6 claude-sonnet-4-6 | Anthropic | $3.60 | $18.00 | 8,192 tokens | 120s |
| Claude Opus 4.8 claude-opus-4-8 | Anthropic | $6.00 | $30.00 | 8,192 tokens | 180s |
| GPT-5.6 Sol gpt-5.6-sol | OpenAI | $6.00 | $36.00 | 8,192 tokens | 180s |
| GPT-5.6 Terra gpt-5.6-terra | OpenAI | $3.00 | $18.00 | 8,192 tokens | 120s |
| GPT-5.6 Luna gpt-5.6-luna | OpenAI | $1.20 | $7.20 | 4,096 tokens | 60s |
| GPT-5.1 gpt-5.1 | OpenAI | $1.50 | $12.00 | 8,192 tokens | 120s |
| GPT-5 mini gpt-5-mini | OpenAI | $0.30 | $2.40 | 4,096 tokens | 60s |
| Gemma 4 26B (Workers AI) default @cf/google/gemma-4-26b-a4b-it | Workers AI | $0.12 | $0.36 | 4,096 tokens | 60s |
Rates are provider list price × 1.2 — the platform margin covers Stripe fees, refunds, and infrastructure. Creators keep 100% of their markup.
How pricing works
- Pay for what a run actually uses. Each run is charged on the tokens the model actually consumed, rounded up to whole credits (minimum 1 credit ≈ $0.0001), plus a small fixed per-job overhead.
- Creator margin is separate and visible. App creators may add a markup of up to 500% on top of the billed rate — it's shown on each app's detail page, and creators keep 100% of it.
- BYOK apps bill a flat 1 credit per run. When a publisher brings their own provider key, their key pays for inference and the platform charges only a 1-credit overhead — markup is forced to zero.
- Spend is capped per job. Every model carries hard caps on input tokens, output tokens, and wall-clock time, so a single run can never overrun its hold.
- No model configured? Apps without a model run on the Workers AI default — cheap, keyless, and never silently billed at premium rates.