| Model | Exact API ID | Input / 1M | Output / 1M | Avg TPS / 24h | Capabilities |
|---|---|---|---|---|---|
| GPT-6.2 AstraOpenAI | gpt-6.2-astra | $0.0672 | $0.252 | Collecting | reasoningcodeflagship |
| GPT-6.2 SolOpenAI | gpt-6.2-sol | $0.0672 | $0.252 | Collecting | reasoningcodegeneral |
| GPT-6.1 SolOpenAI | gpt-6.1-sol | $0.0672 | $0.252 | Collecting | generalcode |
| GPT-6.1 TerraOpenAI | gpt-6.1-terra | $0.0538 | $0.242 | Collecting | fastcode |
| GPT-6 AstraOpenAI | gpt-6-astra | $0.202 | $0.756 | Collecting | reasoningflagship |
| GPT-6 LunaOpenAI | gpt-6-luna | $0.0054 | $0.0202 | 50.3 | fastcode |
| Claude Opus 5.5Anthropic | claude-opus-5-5 | $0.202 | $0.756 | Collecting | reasoningflagship |
| Claude Sonnet 5.5Anthropic | claude-sonnet-5-5 | $0.0672 | $0.252 | Collecting | reasoningcode |
| Claude Haiku 5.5Anthropic | claude-haiku-5-5 | $0.0538 | $0.242 | Collecting | fastcode |
| Gemini 4 HydrogenGoogle | gemini-4-hydrogen | $0.210 | $1.050 | Collecting | reasoningmultimodal |
| DeepSeek V4.1 Flash 1017 UncensoredDeepSeek | deepseek-v4.1-flash-1017-uncensored | $0.104 | $0.312 | Collecting | fastuncensored |
| GLM 5.3 AbliteratedZhipu AI | glm-5.3-abliterated | $0.728 | $2.421 | Collecting | reasoningabliterated |
| GLM 5.3 Flash AbliteratedZhipu AI | glm-5.3-flash-abliterated | $0.218 | $0.910 | Collecting | fastabliterated |
How is a request calculated?
New input, cached input, cache writes, and output tokens are multiplied by their published per-million rates. Reasoning tokens count as output. Your usage log shows the exact tokens and charge for each request.
Why not match every competitor rate?
Some public competitor prices are below our upstream cost. We would rather keep a sustainable margin than quietly lose money, degrade routes, or freeze service. On many directly comparable models SafeAPI is substantially cheaper; on others you can choose the route that makes sense for you.
Do failed requests cost money?
Requests rejected before reaching the upstream are not charged. Once an upstream starts processing, its actual reported token usage can be billable even if a stream disconnects. Estimated or reconciled charges are labeled in your usage history.
How does caching affect my price?
SafeAPI passes through supported provider-side prompt caching. Cache reads and writes use their own rates. We do not promise cache hits or operate a shared semantic cache with your private prompts.