Show HN: 1endpoint – Cheaper access to AI models
An AI API gateway aggregates 15+ LLM models (GLM, GPT, DeepSeek, Gemini, Qwen, Kimi, etc.) under a single compatible endpoint, eliminating the need for per-model integration changes Usage-based pricing starts at $0.0420 per 1M input tokens, with automatic prompt caching reducing costs significantly on repeated conversation turns (e.g., message 12 costs 3.7× less than without cache) The platform supports OpenAI-compatible endpoints (/chat/completions, /responses, /messages) and reports routing ov
Analysis
TL;DR
- An AI API gateway aggregates 15+ LLM models (GLM, GPT, DeepSeek, Gemini, Qwen, Kimi, etc.) under a single compatible endpoint, eliminating the need for per-model integration changes
- Usage-based pricing starts at $0.0420 per 1M input tokens, with automatic prompt caching reducing costs significantly on repeated conversation turns (e.g., message 12 costs 3.7× less than without cache)
- The platform supports OpenAI-compatible endpoints (/chat/completions, /responses, /messages) and reports routing over 748M tokens and 6,505 API requests in a single 24-hour period
- Prompt caching is a core cost-saving feature: cache hits are billed at up to 5× less than cache misses, with blended rates assuming a 70% cache hit ratio for comparison
- A referral program offers 10% credit rewards for referred users who top up, and 1,000 credits equal $1 USD
Why It Matters
This represents the growing trend of AI API aggregation platforms that abstract away model-specific integration complexity, allowing developers to switch between models seamlessly while optimizing costs through intelligent caching. For practitioners, it demonstrates how prompt caching can dramatically reduce per-conversation costs in long-running chat applications, making multi-model orchestration economically viable at scale.
Technical Details
- Multi-model catalog: Supports GLM 5.2/5.3, GPT-5.6 (Luna/Terra/Sol), DeepSeek V4 (Flash/Pro/Vision), Gemini 3.7 Flash, Qwen3.8, Kimi K3, MiniMax M3, Grok 4.6, Opus 5, and Fable 5 — spanning from budget flash models to premium reasoning models
- API compatibility: OpenAI-compatible endpoints including POST /chat/completions, POST /responses, and POST /messages under a single base URL, enabling drop-in model swaps without code changes
- Prompt caching mechanics: Every message in a conversation resend prior context; the gateway charges full price only for new tokens while caching previously seen tokens, with cache hits billed at up to 5× less than misses
- Pricing structure: Input tokens range from $0.0420/M (GLM 5.2) to $1.5000/M (Fable 5); output tokens range from $0.0050/M to $0.1320/M depending on model; blended reference rates assume 70% cache hit and 25% output-to-input ratio
- Credit system: Usage-based billing at 1,000 credits = $1 USD, with spend tracking and a 10% referral reward on topped-up credits
Industry Insight
- The aggressive pricing floor ($0.0420/M input) signals intensifying commoditization of baseline LLM inference, pushing providers to compete on caching efficiency and model quality rather than raw token costs alone
- Prompt caching as a default feature — not a premium add-on — is becoming table stakes for any serious API gateway; developers should architect conversations to maximize cache reuse (e.g., stable system prompts, minimal redundant context)
- The multi-model aggregation model lowers the barrier for A/B testing and fallback routing, suggesting that production systems should treat model selection as a dynamic, cost-aware decision rather than a fixed architectural choice
Disclaimer: The above content is generated by AI and is for reference only.