
ARC-AGI-3 Opus 5 30.2% — Leaderboard & HN Debate
Claude Opus 5 scores 30.2% on ARC-AGI-3 — ~4× GPT-5.6 Sol Max and ~20× Opus 4.8. ARC Prize verified results, cost/task, Fable’s absence, and the HN contamination debate.
Frontier model releases, AI agent guides, Claude Code deep dives, MCP servers, AI courses, careers, workshops, and global AI policy — 600+ articles from the explainx.ai team.

Claude Opus 5 scores 30.2% on ARC-AGI-3 — ~4× GPT-5.6 Sol Max and ~20× Opus 4.8. ARC Prize verified results, cost/task, Fable’s absence, and the HN contamination debate.

Thariq on Claude 5 context engineering: Anthropic cut 80%+ of Claude Code’s system prompt for Opus 5 and Fable 5 with no eval loss. Judgement over rules, progressive disclosure, /doctor.

ClaudeDevs Opus 5 developer guide: Frontier-Bench 43.3%, /claude-api migrate, mid-conversation tool changes, classifier fallbacks, Fast mode ~2.5×, effort defaults in Claude Code.

Anthropic shipped Claude Opus 5 on July 24, 2026: near Fable 5 intelligence at half the price, $5/$25 like Opus 4.8, Frontier-Bench SOTA 43.3%, ARC-AGI-3 30.2%.

Developer am.will used Claude Opus 5 to build a playable Three.js Rocket League clone at ~27% of a 5x Max sub. Opus then drove the car to verify mechanics. Open source + Vercel demo.

Unitree’s AS2-W Super Athlete mixes wheels and legs: 16kg continuous payload, 30+ km unloaded range, RL control. Viral demo, civilian pitch, military speculation — what the robot actually is.

Y Combinator’s Fall 2026 Requests for Startups: 13 ideas from AI tutoring and Army defense tech to multiplayer agents, compute at sea, and self-maintaining APIs. Apply by July 27.

Jul 22, 2026: Amazon confirmed AGI-org cuts after Reuters. Headcount undisclosed; employees cite pretraining, post-training, customization. Nova still a priority.

Jul 22, 2026: AMD equity up to $5B; Anthropic deploys up to 2 GW MI450 Helios (MI455X, Venice, Pensando) from H1 2027. Claude for ROCm; MI355 weekend bring-up.

July 21, 2026: Applied Intuition launches Dana — agentic platform for autonomy, SDV, mining, and robotics. Isuzu and Komatsu early access.

Block’s Buzz (Apache 2.0): self-hostable Nostr relay where humans and agents share channels, git patches, workflows, and audit trails. Desktop, buzz-cli, ACP.

Zenity’s AgentForger (disclosed Jul 2026): one ChatGPT link forged a Workspace Agent under a victim’s identity. OpenAI patched in days — CSRF for agents.

ChatGPT Voice hits macOS/Windows Jul 23, 2026 — GPT-Live full-duplex, Mac Appshots, agent quotas for Plus–Enterprise. Plans, context, vs mobile.

July 21, 2026: Cisco Foundation AI open-weights Antares-350M and Antares-1B for local vulnerability localization, plus a 500-task VLoc Bench.

Anthropic’s Claude Cookbook hit HN: PTC, tool search, compaction, Managed Agents, Fable fallback. What’s worth reading vs what the harness will absorb next quarter.

Accomplish AI’s SharedRoot (Jul 23, 2026): one message escapes Cowork’s Linux VM to host Mac R/W via virtiofs + CVE-2026-46331. Design fixes, not just CVE.

Jul 23, 2026: Claude Voice leaves Haiku-only — Opus/Sonnet on paid, connectors mid-call, 11 languages every plan. Turn-based vs GPT-Live; Free vs paid.

Anthropic’s Jul 24 zoom/crop cookbook: Fable 5 Chartography 29%→73%, Sonnet 5 13%→44%. How 28×28 patches, full-res crops, and cost trade-offs work.
Page 1 of 56 · 1006 articles