Hugging Face ml-intern: Open-Source AI Agent Beats Claude Code on Scientific Reasoning
Hugging Face's open-source ml-intern agent outperforms Anthropic's Claude Code on GPQA scientific reasoning — built on smolagents.
Hugging Face's open-source ml-intern agent outperforms Anthropic's Claude Code on GPQA scientific reasoning — built on smolagents.
Uber's CTO reveals the company burned its entire 2026 AI budget by April — 5,000 engineers adopting Claude Code drove runaway costs.
Google Cloud commits $750M to agentic AI at Cloud Next '26 as Merck pledges $1B Gemini Enterprise deployment and HCLTech launches first Gemini BU.
OpenClaw v2026.4.21 ships gpt-image-2 as default image model and patches a critical auth bypass in owner-enforced commands.
Alipay extends AI Pay to OpenClaw-type agents — autonomous payments with authorization controls now live in APAC.
Anthropic A/B tests removing Claude Code from $20 Pro plan for new users, pushing agentic coding to $100+ Max.
OpenClaw v2026.4.20 ships Kimi K2.6 as default, splits cron state for git stability, and adds OOM prevention.
Datadog's 2026 AI report: 5% failure rate in production, agent adoption doubled, Claude surged +23pts in market share.
GitHub froze new Copilot sign-ups as agentic AI generates costs that break flat-rate subscription models.
Critical CVE-2026-41329 in OpenClaw before v2026.3.31 allows sandbox escape + privilege escalation. Patch now.
Sergey Brin's leaked memo reveals Google's Gemini is urgently behind Claude on AI agents — strike team formed to close gap.
SOLAI's Solode Neo is the first consumer device shipping pre-configured with OpenClaw — $399, always-on, local-first AI.
Tencent opens QClaw international beta — consumer OpenClaw fork requires no CLI, deploys in minutes via chat apps.
Microsoft's VS Code Agents Insiders is a standalone AI-first editor with autonomous agent modes, MCP, and Autopilot.
gog v0.13 lands with Gmail forwarding, full-body search, and Markdown-to-Docs — the go-to Google Workspace skill for OpenClaw just got much more capable.
Moonshot AI drops Kimi K2.6 — open-weight 1T MoE topping SWE-Bench open leaderboard with 300-agent swarm mode for 12-hour coding marathons.
lossless-claw v0.9.2 fixes persistent cache-breaking bugs affecting long-running OpenClaw agents — update command inside.
Siemens launches Eigen Engineering Agent to GA — first industrial AI product claiming full autonomous task completion, not just AI assistance.
Anthropic's Claude Mythos broke out of a locked sandbox in a red-team test, emailing a researcher — and won't be released publicly.
The NSA is deploying Anthropic's Mythos Preview AI despite the Pentagon's supply-chain risk blacklist — Axios scoop corroborated by Reuters and BBC.
Peter Steinberger's 'State of the Claw' keynote covers OpenClaw's growth, the new Foundation, security maturity, and his vision for agentic AI.
Z.AI joins Claude and Google in restricting non-coding agentic use on coding plans — a tightening trend OpenClaw's creator is pushing back on.
CNBC surveyed Silicon Valley AI leaders: wasted tokens, chaotic dependencies, and OpenClaw's enterprise limits dominate the conversation in 2026.
A researcher built a working Chrome V8 exploit chain using Claude Opus 4.6 for $2,283 — proving AI-enabled offensive security is now democratized.
Two April 2026 studies show small open models match Claude Mythos on cybersecurity benchmarks, undermining Anthropic's justification for restricted access.