Speko open-sources a voice-AI router that auto-picks your STT+LLM+TTS
MIT-licensed and self-hostable with your own keys, it benchmarks a dozen vendors per stage. Also: ChatGPT for Teens auto-enrolls under-18s; Sol hits 750 tok/s.
Copy markdown
One endpoint picks your whole voice stack
Speko benchmarks about a dozen speech-to-text, LLM, and text-to-speech vendors per category and auto-selects the best combo for your latency, cost, and language limits — with automatic failover if a provider refuses mid-connect. You point one endpoint at it instead of hand-wiring three vendors.
Self-host it free, bring your own keys
The gateway is MIT-licensed and free to run indefinitely with BYOK; you only pay for Speko's hosted routing and managed keys. Public boards at benchmarks.speko.ai rank STT, LLM, TTS, turn-taking, and speech-to-speech — including cases where rivals beat Speko's own picks.
ChatGPT for Teens auto-enrolls under-18s
OpenAI launched a 13-17 tier that blocks self-harm and sexual content and uses age prediction to route suspected minors into it automatically, with parental quiet-hours and high-risk alerts. If you build on ChatGPT surfaces, platform-level age gating is now live.
GPT-5.6 Sol hits 750 tokens/sec on Cerebras
OpenAI's Ultrafast preview runs GPT-5.6 Sol up to 14x faster (~750 output tokens/sec) on Cerebras silicon with no quality drop; pricing is still unannounced and access is invite-only. It changes what real-time agents and interactive tools can feel like.