AssemblyAI releases Universal-3.5 Pro speech model

// 1h agoMODEL RELEASE

AssemblyAI releases Universal-3.5 Pro speech model

Universal-3.5 Pro is AssemblyAI's most accurate speech-to-text model, now available through their Realtime and Async endpoints. The model natively supports code-switching across 18 languages, features improved speaker diarization, and supports contextual prompting for domain-specific terminology.

// ANALYSIS

AssemblyAI's focus on native code-switching is a massive step forward for building international voice agents, rendering single-language constraints obsolete.

–Real-time code-switching across 18 languages facilitates natural multi-lingual conversations without requiring separate pipelines.
–Enhanced diarization directly improves the downstream utility of transcripts for meeting summaries, CRM logging, and QA.
–The combination of low latency (Realtime endpoint) and high accuracy is critical for next-generation conversational AI interfaces.

// TAGS

sttassemblyaiartificial-intelligenceapidevtool

DISCOVERED

1h ago

2026-07-08

PUBLISHED

7h ago

2026-07-08

RELEVANCE

8/ 10

AUTHOR

[REDACTED]

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

VIDEO7m ago

HASE co-evolves model weights, harness

Harness-Aware Self-Evolving (HASE) is an agentic reinforcement learning framework that allows a single model to co-evolve its policy weights, task solutions, and environment harness in a unified multi-turn action space. By enabling the model to dynamically modify harness components like prompts, memory formatting, and validators, HASE allows smaller models like Qwen3-8B to match or beat the performance of models as large as 120B parameters in domains like text classification and alpha factor mining.

RESEARCH7m ago

GaP boosts robot policy reliability

Graph-as-Policy (GaP) is a robotics framework designed to bridge the reliability gap in variational automation tasks by representing robot policies as directed computation graphs of modular skills. Developed by researchers from UC Berkeley, NVIDIA, CMU, and Bosch, the framework uses an LLM-in-the-loop multi-agent system to construct, test, and refine these graphs in simulation before deployment.

NEWS7m ago

Lilian Weng outlines harness engineering for RSI

Lilian Weng argues that recursive self-improvement (RSI) in AI lies in optimizing the "harness"—the orchestration layer surrounding a base model—rather than directly rewriting weights. Weng details how harness engineering is transitioning into code-based meta-optimization, though full RSI remains bottlenecked by evaluation and memory limitations.