Report: Rime Prism lets you rewrite a voice agent's reply mid-call
The speech-to-speech model reportedly reaches first audio in 0.57s — plus cheaper near-frontier models out of China and Claude's new legal workspaces.
Copy markdown
Edit the words before your agent speaks them
Rime reportedly launched Prism, a speech-to-speech model that reaches first audio in ~0.57s and exposes a text checkpoint — your app can read or rewrite the generated reply before the caller ever hears it. It's fine-tunable and runs in Rime's cloud, your own VPC, or on-prem. For voice-agent builders, that's real guardrails and brand-voice control without trading away low latency.
Near-frontier web agents at a fraction of the price
A report says DeepSeek's V4.1-Flash and Moonshot's Kimi K3.1 Pro were distilled from OpenAI's GPT-6 Astra. DeepSeek reportedly reaches a 75.0 browser-operation score (Astra: 77.9) at about $0.22 per call versus Astra's $10.77 — roughly 50x cheaper. If the numbers hold, agentic browsing tasks you shelved on cost are suddenly worth building.
Claude ships ready-made legal workspaces
Anthropic reportedly rolled out prebuilt Claude workspaces for contract review, legal citation checking, and tracking regulatory changes: start from a sample, connect your own documents, and share the workspace with your team. Even outside law, it's a clone-and-adapt template for standing up a domain workflow instead of wiring prompts and connectors from scratch.