Models and Pricing
CodeGate bills per token, in prepaid credits. Prices are quoted per 1 million tokens (1M). The headline discount for Anthropic models is 85% off list.
How pricing works
Every request is billed on four counters:
- Input tokens, everything you send: system prompt, prior messages, tool definitions.
- Cache-write tokens, tokens written into a prompt cache. Slightly more expensive than input, one-time.
- Cache-hit tokens, tokens fetched from a warm cache on a subsequent request. Cheap.
- Output tokens, everything the model returns.
Rates below are per 1M tokens.
Anthropic (Claude)
Every Claude row is 85% off Anthropic's public list price.
| Model | Input | Cache write | Cache hit | Output |
|---|---|---|---|---|
| Claude Fable 5 | $1.00 | $1.25 | $0.10 | $5.00 |
| Claude Opus 4.8 | $0.50 | $0.63 | $0.05 | $2.50 |
| Claude Sonnet 5 | $0.30 | $0.38 | $0.03 | $1.50 |
| Claude Haiku 4.5 | $0.08 | $0.10 | $0.008 | $0.40 |
Model IDs to send in the request body: claude-fable-5, claude-opus-4-8, claude-sonnet-5, claude-haiku-4-5-20251001.
OpenAI
| Model | Input | Cache hit | Output |
|---|---|---|---|
| GPT-5.5 | $0.50 | $0.05 | $3.00 |
Model ID: gpt-5.5.
What is not shipped yet
- Gemini. Planned. Not currently routable. When it lands the pricing will appear on this page.
- Embeddings. Not shipped.
- Image generation. Coming soon. Endpoint will be
POST /v1/images/generations. - Audio (Whisper transcription, TTS). Not on the roadmap.
How the 85% discount works
CodeGate runs its own pool and infrastructure. Requests hit real Anthropic models and return real Anthropic responses; what you save is passed through from that pool.
Cache pricing details
Anthropic's prompt caching pays back after roughly two hits: the first request pays cache-write; every subsequent request that shares that prompt prefix pays cache-hit instead of input. On big system prompts (long tool definitions, retrieval snippets) the savings compound quickly.
To use caching, structure your requests so the shared prefix is at the front, and enable caching flags per Anthropic's caching docs. CodeGate passes those flags through unchanged.
Custom pricing
The 85% discount is already priced in. Larger accounts (typically $5,000+/month) can request a custom quote. Join the Discord with your usage and we will get back to you.
Where to find balance and spend
- Per request: Console, Logs. Every call shows the model, token counts, and cost.
- Per key: Console, API Keys shows lifetime spend per token.
Next
- Billing and Quotas for how top-ups work.
- Rate Limits.