On-Demand
Free
- Router access with LLM usage at provider cost, up to 70 minutes of TTS, built for evaluation and prototyping
Inworld Router: Pricing, Features, Pros & Cons
Provider rates with no markup on routed models, inside Inworld platform plans from free to custom Enterprise
What is InWorld AI?
Inworld Router, which Inworld also calls the Realtime Router, is an LLM router from Inworld AI, the realtime voice and model company founded in 2021. One endpoint compatible with both the OpenAI and Anthropic SDKs reaches 220+ models, and routed third-party models bill at the provider's rate with no markup, including on your own keys.
Routing is rule-based and built around experiments, which sets it apart from LLM routers that pick a model from the prompt alone. Conditions written in CEL route on any request metadata, such as language, plan tier, or detected emotion, and traffic can be split between models with sticky user assignment for A/B tests. Because Inworld also builds voice models, a single chat completions request can return both text and synthesized speech.
Observability is request-level. Each call logs the model chosen, the attempt chain, time to first token, cost, and the routing reason, and that data can stream to Mixpanel or BigQuery. Inworld's own guidance for A/B tests is to measure quality in your own analytics, and the product has no built-in evals or scoring.
Key features
Core capabilities this platform advertises.
- Conditional routing with CEL expressions on request metadata
- A/B testing with weighted splits and sticky user assignment
- Text and TTS speech from one API call
- TTFT-based automatic failover
Strengths and tradeoffs
What this tool does well, and the limitations to keep in mind.
Pros
- No markup on routed third-party models, including bring-your-own-key traffic
- Routing rules change without a redeploy
- Sticky user assignment keeps A/B tests consistent across sessions
Cons
- No built-in evals, so quality per model has to be measured in your own analytics
- Observability is request-level, not nested agent traces
Plans & pricing
What's included in each plan, and how the tiers compare.
Creator
$25
Per month
- Content creation and small projects; LLM usage at provider cost
Builder
$100
Per month
Common use cases
Consumer AI apps, voice agents, and AI companions that need to route by user, test models on live traffic, and add speech output
- Routing free and paid users to different models
- A/B testing two models on live users before a full rollout
- Building voice agents that need LLM output and speech together
Using InWorld AI with Respan
Respan is the AI router with observability and automated evals built in. When you compare one model against another, Respan scores each model's output on live traffic and links every score to the trace behind it, so you measure quality directly instead of inferring it from downstream analytics.
- One API for 1,000+ models, and switching models takes a one-word change
- Automatic fallbacks when a provider errors or rate-limits
- Evals on live production traffic, sliced by model
- Experiments that test a model change against real cases before rollout
- Nested traces across every LLM call, tool run, and agent turn
- Costs broken down by model, request, and end customer
Compare models on quality, not just cost
10KFree traces/mo
1,000+Models
Best InWorld AI alternatives & competitors
Top companies in LLM Gateways you can use instead of InWorld AI.
Compare InWorld AI
Side-by-side comparisons with other tools in this category.
Best integrations for InWorld AI
Companies from adjacent layers in the AI stack that work well with InWorld AI.
