AI Daily Brief — 22 April 2026
Google Cloud Next '26 opens — Gemini Enterprise Agent Platform debuts; 8th-gen TPUs ship as TPU 8t for training (9,600-chip superpod) and TPU 8i for inference…
Google Cloud Next '26 opens — Gemini Enterprise Agent Platform debuts; 8th-gen TPUs ship as TPU 8t for training (9,600-chip superpod) and TPU 8i for inference…
Adobe Summit day-two keynote launches CX Enterprise and Brand Intelligence; NVIDIA's Jensen Huang joins Shantanu Narayen on stage.
NVIDIA + Adobe + WPP announce agentic AI for enterprise marketing. The CX Enterprise Coworker, powered by NVIDIA Agent Toolkit, ahead of Adobe Summit day-two demo.
Quiet Sunday — industry pause before Google Cloud Next '26 (April 22-24) and the next wave of agentic-AI releases.
Saturday digest of the Anthropic Opus 4.7 + Claude Design week. Coverage circles back to Google DeepMind's Gemini Robotics-ER 1.6, which shipped four days earlier with…
Anthropic Labs launches Claude Design — collaborate with Claude to ship polished UI, slides, and one-pagers, powered by Opus 4.7 and trained on your team's design…
Anthropic ships Claude Opus 4.7 — +10.9 points on SWE-bench Pro vs 4.6, higher-res vision, file-system memory, new xhigh reasoning tier — and openly concedes Mythos…
Stanford HAI publishes the 2026 AI Index — frontier models gain 30 points on Humanity's Last Exam; Anthropic leads by 2.7%. Dwarkesh hosts Jensen Huang.
Midjourney releases V8.1 Alpha — restoring V7-style aesthetics on V8.0's technical base, with faster HD mode, prompt shortener and updated Describe.
A quiet Monday — no major model drop. Industry watch focuses on Mira Murati's Thinking Machines Lab and prediction-market pricing for the next frontier release.
MiniMax open-sources M2.7 — a self-evolving agent model trained with self-play — while Ollama ships a stability release patching Gemma 4 and GLM-5.1 crashes.
OpenAI's Codex CLI gains voice control (Realtime V2) and Model Context Protocol support, matching capabilities previously limited to Claude-side clients.
Coverage centres on Mythos's hacking capabilities being deliberately restricted; OpenAI projects $100B ad revenue by 2030; GLM-5.1 keeps trending.
Western coverage processes Meta's Muse Spark closed-source pivot and Zhipu's GLM-5.1 MIT release — a single Wednesday that reshaped the open-vs-closed lines.
Meta debuts Muse Spark — closed-source, multimodal, MSL's first model. Z.ai open-sources GLM-5.1 with an MIT licence; Zhipu stock surges 15.9%.
Anthropic unveils Claude Mythos Preview — a frontier model a full capability tier above Opus 4.7 — and Project Glasswing with 12 founding orgs and ~40…
Cognition's Windsurf ships an Adaptive model router with pricing transparency. Devin Desktop adds NVIDIA multi-agent support.
Tufts researchers unveil a neuro-symbolic Visual-Language-Action system blending statistical pattern recognition with symbolic reasoning for robotics.
Quiet Saturday digesting Gemma 4 and Cursor 3 — community write-ups highlight Gemma's clean Apache 2.0 licence and top-three Arena AI placement.
Gemma 4 benchmarks land: the 31B Dense scores 89.2% on AIME 2026 and ranks top-three globally on Arena AI among open models.
Cursor 3 ships an agent-first IDE with eight parallel agents on Composer 2 (built on Kimi K2.5). Google releases Gemma 4 — four open multimodal variants…
Oracle terminates 30,000 employees to fund a $50B AI-infra pivot. Perplexity makes Computer in Slack public. Amazon Science publishes on LLM-TTS robustness.
Oracle executes the largest workforce reduction in its 49-year history — ~30,000 employees cut via 6 a.m. emails to free $8–10B annual cash flow for AI…
Monday: industry watches Q1 wrap reports. Oracle's expected restructuring announcement looms; Q1 2026 venture funding numbers point to a record-breaking AI quarter.