ByteSpike vs aireiter and b.ai: how the gateway choice changes
Three model gateways serve overlapping markets. The differences are concrete: how you authenticate, how billing works, which models are first-class, and what happens when calls fail. Here's what we'd say if we were the customer.
We respect the work both aireiter and b.ai have shipped. They built the category, and the category itself is what makes ByteSpike viable. This post isn't about who's better in the abstract — it's about the concrete decisions a developer makes when picking an aggregator, and where ByteSpike took a different turn.
Authentication shape
aireiter uses a custom header scheme. b.ai uses Bearer with an llmservice prefix. ByteSpike accepts the native auth shape of whichever protocol you're calling: `x-api-key` for Anthropic Messages, `Authorization: Bearer` for OpenAI Chat Completions and Responses. Same key, two header layouts, no SDK forking.
Pricing transparency
aireiter publishes a tier table with bands. b.ai publishes per-model rates inside the dashboard, behind login. ByteSpike publishes the full 45-model rate card publicly at docs.bytespike.ai/pricing — input / cache-write / cache-read / output for every chat model, per-image and per-second rates for multimodal, refreshed nightly. There's no markup paragraph, and you can hit GET /api/pricing programmatically for the same data.
Failure billing
aireiter bills 5xx at half rate by default. b.ai's policy varies by upstream provider. ByteSpike's policy is one line: any non-2xx response is free, period. The narrow exception is video tasks cancelled after the GPU starts rendering — partial seconds bill, because the work was already done.
Chinese-market integration
aireiter focuses on US-market models. b.ai supports Chinese vendors but with English-only billing. ByteSpike ships native CJK across the entire surface: WeChat Pay, Alipay, CNY-denominated subscriptions, Chinese-tuned models (Kimi K2.6, GLM-5.1, MiniMax M2.7) as first-class citizens with the same shape and same billing as their Western counterparts.
When to pick which
- **Pick aireiter** when you want US-market models only and care more about the Discord community than billing transparency.
- **Pick b.ai** when you've already integrated and the migration cost outweighs the policy differences.
- **Pick ByteSpike** when transparent rate cards, native CJK payment, and a `failures-don't-bill` line item all matter — or you'd rather write your code against the Anthropic protocol and call any model with the same body shape.