AssemblyAI 在 2026年5月 发布的内容

AssemblyAI Universal-3 Pro Streaming gets punctuation-based turn detection and multilingual code-switching

AssemblyAI updated its Universal-3 Pro Streaming STT model with punctuation-based turn detection (replacing confidence thresholds) and native multilingual code-switching across six languages. The mode…

AssemblyAI adds fallback models to LLM Gateway for resilient API calls

AssemblyAI’s LLM Gateway now supports automatic fallback to alternative models if the primary model fails, with up to two fallback tiers. Developers can specify fallback models in API requests, overri…

AssemblyAI Voice Agent API adds secure browser integration with WebSocket support

AssemblyAI introduced a secure browser integration for its Voice Agent API, enabling developers to connect directly from client-side code without exposing API keys. The update includes a token-based W…

AssemblyAI integrates with Recall.ai for Zoom real-time transcription

AssemblyAI launched a new integration with Recall.ai to enable real-time Zoom meeting transcription using AssemblyAI's speech-to-text models. The solution requires configuring AssemblyAI as a transcri…

AssemblyAI launches LLM Gateway with 25+ models and EU data residency

AssemblyAI introduced the LLM Gateway, a unified API providing access to over 25 models from providers like Claude, GPT, and Gemini through a single interface. The service supports US and EU data resi…

AssemblyAI adds real-time PII redaction to streaming transcripts

AssemblyAI introduced streaming PII redaction for its real-time speech-to-text API, automatically detecting and removing personally identifiable information from final transcript turns before delivery…

AssemblyAI launches Medical Mode for medical terminology transcription

AssemblyAI introduced Medical Mode, a domain-specific feature that improves transcription accuracy for medical terminology such as medications, procedures, conditions, and dosages. It supports English…

AssemblyAI launches Universal-3 Pro Streaming for real-time voice agents

AssemblyAI introduced Universal-3 Pro Streaming, a real-time speech model optimized for sub-10-second utterances with native multilingual code switching and entity accuracy. It offers sub-300ms latenc…

AssemblyAI launches Universal-3 Pro for high-accuracy voice transcription

AssemblyAI introduced Universal-3 Pro, its most advanced Voice AI model, optimized for accuracy with entities, rare words, and domain-specific terms while maintaining speed. It supports both pre-recor…

AssemblyAI LLM Gateway adds structured outputs with JSON schema support

AssemblyAI’s LLM Gateway now supports structured outputs, allowing developers to constrain model responses to specific JSON schemas for reliable parsing. The feature uses the `response_format` paramet…

PII Redaction

AssemblyAI expanded its PII redaction feature to include redacted audio file generation (with beep or silence options) and the ability to return both redacted and unredacted transcripts in a single AP…

Multichannel Transcription

AssemblyAI now supports per-channel diarization in multichannel audio transcription, allowing separate speaker labeling within each channel. This feature increases transcription time by ~40% and appli…

Structured Outputs

AssemblyAI’s LLM Gateway now supports structured outputs, allowing users to constrain model responses to a specific JSON schema for predictable parsing. The feature includes automatic JSON repair for …

AssemblyAI expands speech-to-text models with Universal-3 Pro and Medical Mode

AssemblyAI introduced Universal-3 Pro, a high-accuracy speech-to-text model supporting 6 languages with advanced prompting and code switching, alongside Universal-2 for 99 languages. They also launche…

AssemblyAI launches evaluation tools for Speech-to-text model selection

AssemblyAI introduced evaluation tools to help customers select the best Speech-to-text model for their use cases by running assessments on their own audio data. The tools measure accuracy (WER, FWER)…

AssemblyAI updates AI coding agent integration with MCP server and agent instructions

AssemblyAI released updated integration prompts and an MCP server to help AI coding agents (Cursor, Claude Code, Copilot, Devin) fetch current documentation and avoid stale training data. The setup in…

How Super scaled real-time voice agents for real estate with AssemblyAI

Super, a vertical Voice AI platform for real estate, integrated AssemblyAI’s Universal-3 Pro Streaming to improve turn detection, keyterm accuracy, and reduce STT costs by ~30%. The integration enable…

Pricing change detected for Assemblyai

Pricing updated for Assemblyai: - Pay-as-you-go: Custom pricing → $0.21/mo - Pay-as-you-go: new promotion — Free to start. $0.21/hr after that – no contracts, no minimums, no credit card required. - R…

Pricing updated for Assemblyai: - Plan features or structure changed