Speko
Speko is a multilingual voice AI routing and infrastructure platform. It provides a hosted, provider-neutral data plane for speech-to-text, LLM, text-to-speech, and speech-to-speech, exposing a single API and typed contracts. Speko benchmarks 56 speech and language models across 10 languages, and routes each session to the model that performs best for the user's language, rather than relying on English-only rankings. Speko's Router is hosted at router.speko.dev and handles managed routing and provider connectivity. The platform also offers an open Gateway runtime that can run alongside customer applications, enabling provider-direct streaming, local BYOK credentials, and optionally managed routes. Native integrations are available for LiveKit, Pipecat, and MCP, allowing voice agent developers to plug Speko into existing frameworks with a few lines of code. Speko's public model comparison covers accuracy (WER) and cost per minute across the major model stages, with language-specific benchmark coverage for speech-to-text and text-to-speech. The company is backed by Y Combinator and provides a console, API key access, documentation, a status page, and a changelog. Specific pricing is not publicly listed on the homepage.
- Category
- AI API
- Pricing
- Unknown
Problem solved
Developers building voice AI applications must choose among many speech and language models, but most benchmarks are English-only and model quality varies by language. Speko solves this by measuring model performance per language and routing each session to the strongest model, so teams can offer multilingual voice experiences without manually benchmarking and integrating multiple providers.
Use cases
- Building multilingual voice assistants and agents that adapt to the user's language
- Routing speech-to-text, LLM, and text-to-speech calls through a single API
- Comparing model accuracy and cost across languages before selecting a model
- Running provider-direct voice pipelines with local credentials via the Gateway
- Integrating voice AI into LiveKit or Pipecat based applications
- Using MCP to add voice AI routing to Claude or Cursor workflows
- Optimizing speech and language model spend per session based on cost/accuracy trade-offs
Audience
- Developers
- ML Engineers
- Startups
- Enterprise
Pros
- Benchmarks speech and language models across 10 languages, exposing language-specific performance.
- Single hosted API routes each session to the best-performing model for the user's language.
- Includes an open Gateway for provider-direct streaming with local BYOK credentials.
- Provides cost and accuracy comparisons across STT, LLM, TTS, and speech-to-speech stages.
- Offers native integrations with LiveKit, Pipecat, and MCP for common voice agent frameworks.
Cons
- Pricing is not listed publicly, making cost estimation harder before sign-up.
- Benchmark coverage is limited to 10 languages, which may not cover all user groups.
- The core routing service is hosted, so applications depend on Speko's infrastructure availability.
- No self-hosted option is offered for the full Router, though the Gateway is open.
Official website: https://speko.ai/
Canonical LambdaBase page: https://www.lambdabase.com/tools/speko