Skip to content
Source · Daily Brief

AI Daily Brief — 23 July 2026

Three leading labs each shipped consumer-facing products on the same day — Anthropic upgraded Claude’s voice mode to its most capable models, OpenAI brought ChatGPT Voice to the desktop with computer-control capabilities and launched ChatGPT Health for all U.S. users, and xAI rolled out Grok 4.5 across all platforms — while AMD disclosed a commitment of up to $5 billion in an infrastructure deal with Anthropic.

Top stories

  • Claude voice mode reaches Opus and Sonnet. Until today, Claude’s voice mode was limited to the lighter Haiku model. Anthropic announced that Opus and Sonnet are now available in voice, with integrations into Gmail, Slack, and Canva extending the reach further. via TechCrunch, via The Verge
  • ChatGPT Voice lands on the desktop with computer control. OpenAI rolled out the feature globally to Plus, Pro, Business, and Education subscribers on macOS and Windows. Powered by GPT-Live, it can listen, speak, and coordinate multiple agents running in ChatGPT Work or Codex simultaneously. via X · @OpenAI
  • ChatGPT Health now available to all U.S. users. OpenAI connected the chatbot to Apple Health, medical records, and wellness services for all American users. Free users receive responses from an older model; paying subscribers access the more capable GPT-5.6 Sol. via TechCrunch
  • Grok 4.5 ships to all platforms. xAI made Grok 4.5 available on grok.com, X, and iOS/Android apps. Elon Musk described it as fast, token-efficient, and near Fable-level quality at a fraction of the cost. Separately, a Capy agent running on Grok 4.5 reportedly refuted a graph-theory conjecture open for roughly 30 years. via X · @elonmusk
  • Flux 3 generates video with native audio. Black Forest Labs released Flux 3, its first multimodal model producing videos up to 20 seconds with synchronized native sound. The company’s own benchmarks place it slightly ahead of Seedance 2.0; independent evaluations are not yet available. via THE DECODER
  • Etched reaches a $10.3 billion valuation. The AI inference chip startup, backed by Sequoia, hit the figure on claims that its architecture accelerates any AI model without requiring GPUs, with shipments to customers expected later in 2026. via TechCrunch

Who shipped

OpenAI had the heaviest product day, launching ChatGPT Voice on desktop and ChatGPT Health in the same cycle. Anthropic matched the pace with voice-model upgrades extending Claude to its most capable tiers, backed by a large AMD hardware commitment. xAI completed a platform-wide Grok 4.5 rollout, and Black Forest Labs crossed a new threshold in generative media with audio-native video output in Flux 3.

Open-source pulse

Poolside released Laguna S 2.1, a 118B MoE coding model trained to revise failed approaches and persist through long agentic sessions; the company claims it beats several larger open-weight rivals. Andrew Ng released OpenWorker under the MIT licence — a desktop AI agent that delivers finished documents and completed tasks rather than chat responses, supporting both API-hosted and local Ollama models. Kimi K3 from Moonshot AI is set to arrive on Together AI on 27 July, with early comparisons placing it near Mythos Preview on coding tasks.

Money, infra & hardware

AMD agreed to invest up to $5 billion in Anthropic under an infrastructure agreement; Anthropic will deploy up to two gigawatts of AMD Instinct MI450-series compute, with the first gigawatt targeted for the first half of 2027 (via AI News). Security startup AegisAI raised $36M into AegisAI in a Battery Ventures-led Series A to block AI-driven spear phishing. Alphabet separately reported its first-ever negative quarterly cash flow owing to AI infrastructure spending, even as Google Cloud grew 82 percent year-over-year; CEO Sundar Pichai confirmed a Gemini 4 training run has begun.

Quiet corners

Meta appeared only in coverage of a misfired AI optimism advertisement scored to a David Bowie song about human extinction. Mistral and Cohere were absent from today’s cycle.

By the numbers

  • 399 stories published today across 40+ sources
  • Most-mentioned model: GPT (ChatGPT Voice, Health, ongoing HuggingFace incident fallout)
  • Most-mentioned lab: OpenAI
  • Notable absences: Meta, Mistral, Cohere

Compiled by AI Feed’s editor from all 399 headlines published on 23 July 2026.