Ce que AssemblyAI a publié en juin 2026

AssemblyAI adds HTTP tools to Voice Agent API for server-side requests

AssemblyAI introduced HTTP tools for its Voice Agent API, allowing agents to execute server-side HTTP requests directly without client-side round trips. Agents can now define HTTP endpoints with JSON …

AssemblyAI launches Voice Agent API for reusable voice agents

AssemblyAI introduced a Voice Agent API enabling users to configure a reusable voice agent once via a single REST call, storing its personality, voice, and tools on their servers. The agent can then b…

Universal-3.5 Pro Realtime: the first streaming STT model that takes the agent's question as input

AssemblyAI released Universal-3.5 Pro Realtime, a flagship realtime speech-to-text model featuring conversation context carryover and support for 18 languages at flagship accuracy. The model reduces w…

AssemblyAI adds PII redaction to transcriptions with hash and entity_name substitutions

AssemblyAI introduced a PII Redaction model that automatically identifies and removes sensitive personal information from transcriptions, replacing it with either hashed values (####) or entity placeh…

From PLG Pains to AI Gains: Rethinking Retention

From PLG Pains to AI Gains: Rethinking Retention Churn Insights Read more →

Dylan Fox on promptable Speech Language Models & the future of voice AI

Dylan Fox on promptable Speech Language Models & the future of voice AI Voice-ai-newsletter News Read more →

Jena Dunay and Meg Colón on AssemblyAI's remote-first culture.

Jena Dunay and Meg Colón on AssemblyAI's remote-first culture. Open Insights Read more →

AssemblyAI adds contextual prompting and keyterms for streaming transcription

AssemblyAI introduced two new parameters—`prompt` for contextual descriptions and `keyterms_prompt` for explicit term lists—to improve streaming transcription accuracy. Contextual prompts can be domai…

AssemblyAI adds Universal-3 Pro Streaming model for real-time transcription

AssemblyAI introduced a new top-tier streaming speech model, Universal-3 Pro Streaming (u3-rt-pro), offering best-in-class accuracy with native multilingual code-switching and advanced prompting. Two …

AssemblyAI adds EU data residency and global routing to LLM Gateway

AssemblyAI introduced an EU endpoint for its LLM Gateway to ensure data stays within the European Union, supporting Anthropic Claude and Google Gemini models. A global routing option was also added to…

AssemblyAI adds contextual and keyterms prompting to Universal-3 Pro Streaming

AssemblyAI introduced two new prompting methods for Universal-3 Pro Streaming: contextual prompting (natural-language descriptions of audio) and keyterms prompting (explicit term lists). Contextual pr…

AssemblyAI adds real-time speaker diarization and multichannel streaming

AssemblyAI’s Streaming API now supports real-time speaker diarization and multichannel transcription via a new `speaker_labels: true` parameter. Each turn and word in the transcript includes a speaker…

Brex Benchmark May 2025: AI and SaaS spending trends among startups and enterprises

Brex’s May 2025 Benchmark reveals Anthropic leads AI spending among startups (2x OpenAI) while OpenAI dominates enterprise AI spend (4x Anthropic). Cursor, Pinecone, and Replit are fast-growing AI too…

AssemblyAI introduced a PII Redaction model that automatically identifies and removes sensitive personal information from transcripts, replacing it with either hashed values (e.g., ####) or entity pla

AssemblyAI adds browser integration for Voice Agent API

AssemblyAI introduced a browser integration for its Voice Agent API, enabling direct WebSocket connections from client-side apps using temporary tokens. The API key remains server-side only, with toke…

AssemblyAI LLM Gateway adds fallback model support for resilient LLM calls

AssemblyAI’s LLM Gateway now supports specifying fallback models to automatically retry requests if the primary model fails, with up to two fallback tiers and configurable retry behavior. Fallback mod…

AssemblyAI adds real-time PII redaction for streaming transcripts

AssemblyAI introduced streaming PII redaction to automatically detect and redact personally identifiable information from streaming transcripts in real time. The feature applies only to final turns an…

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

AssemblyAI introduced Universal-3 Pro Streaming with sub-300ms latency, formatted output, and punctuation-based turn detection for voice agents. The model now supports native multilingual code-switchi…

AssemblyAI adds Universal-3 Pro and model routing for pre-recorded audio

AssemblyAI introduced a new `speech_models` parameter for pre-recorded audio transcription, allowing users to specify model priority. The system defaults to `universal-3-pro` and `universal-2` if omit…