Live Feed
Thinking Machines Lab Ships Inkling: 975B Open-Weight MoE, 77.6% SWE-bench, Enters Agent Arena
Mira Murati's Thinking Machines Lab releases Inkling — a 975B Mixture-of-Experts model trained on 45 trillion tokens across text, image, audio, and video, available Apache 2.0 with full weights. Architecture mirrors DeepSeek-V3. Added to Text, Code, and Agent Arena leaderboards within five days of launch.
Hut 8 Seals Second $9.8B Lease to Fully Commercialize 1GW Beacon Point at $19.6B
The same investment-grade tenant that signed Phase 1 has doubled its contracted capacity to 704 MW with a second 15-year lease, filling the entire 1GW Beacon Point campus. Hut 8's total portfolio aggregate base-term contract value reaches $26.6 billion.
GPT-5.6 Found a $500K WordPress RCE for $25 — 500 Million Sites Were Exposed
A security researcher used GPT-5.6 and $25 in API credits to find wp2shell, twin pre-authentication RCE flaws in WordPress Core. Every 6.9 and 7.0 site was vulnerable until emergency patches shipped. Public exploits are now circulating.
US Power Companies Are Using Eminent Domain to Seize Land for AI Data Center Infrastructure
Power utilities in Georgia and other states are invoking eminent domain — the government's power to compel land sales — to build transmission lines serving AI data centers. Homeowners facing forced sales call it theft. Georgia Power says it's a last resort used in under 1% of cases.
Context Is the Hidden Layer: 300-Run Study Shows Agent Setup Quality Predicts Failure Before the Agent Runs
A new paper from ProofAgent validates context-engineering quality as a leading indicator of AI agent reliability across 300 multi-turn evaluations and 7,500 turns. The weakest context is often the cheapest — and the most dangerous.
Claude Fable Helps Disprove the 85-Year-Old Jacobian Conjecture — Verified Arithmetic, One Tweet
Levent Alpoge, a number theorist at Anthropic, used Claude Fable 5 to produce an explicit counterexample to the Jacobian Conjecture — an open problem since 1939 and one of Smale's 23 hardest questions for the 21st century. The counterexample is three lines of arithmetic anyone can verify by hand.
Google's TPU Push Into Neoclouds Is Hitting Nvidia's Distribution Moat — CoreWeave, Nebius, and Lambda Won't Switch
Google is trying to sell TPUs through neocloud providers to compete with Nvidia in the market that sits between hyperscalers and AI labs. The Information reports CoreWeave, Nebius, and Lambda have said no. Smaller providers and a Blackstone vehicle are the fallback.
AI Access Collapses 'I Don't Know' Rate From 44% to 3% — Accuracy Drops by Two-Thirds, Confidence Doubles
A three-university controlled study gave participants access to an AI that was deliberately wrong on the test questions. Accuracy fell from 27% to 9%. Confidence climbed from 30% to 76%. Monetary incentives barely moved the numbers.
GPT-5.5-Cyber Hits 85.6% CyberGym as OpenAI's Daybreak Expands to Patch the Planet
OpenAI has moved GPT-5.5-Cyber from limited preview to a broader trusted-defender release at 85.6% CyberGym, 4 points above the base model. A new Patch the Planet initiative with Trail of Bits, HackerOne, and Calif will deliver AI-assisted fixes directly to cURL, Python, Go, Sigstore, and five other critical open-source foundations.
OpenAI Cuts Codex Context Window 27%, From 372k to 272k Tokens
A merged pull request in the Codex repository quietly reduced the model context window from 372,000 to 272,000 tokens in v0.144. Developers noticed immediately: compaction fires earlier, long sessions degrade faster, and the gap with Claude Fable 5's extended context just got wider.
Alibaba Previews Qwen 3.8: 2.4T Parameters, Open-Weight Release Ahead, Claims Second Only to Fable 5
Alibaba launched Qwen3.8-Max-Preview on July 19, positioning its 2.4-trillion-parameter flagship as trailing only Claude Fable 5. No independent benchmarks yet, but open weights are coming. The Chinese open-model arms race now runs at trillion-parameter scale.
Fable 5 Tops Arena's New Image-to-WebDev Leaderboard at 1627 — Anthropic Occupies All Top 7
Arena's Image-to-WebDev leaderboard, which added Claude Fable 5 and Sonnet 5 High on July 17, shows Anthropic holding every position from #1 through #7. GPT-5.5-xhigh enters at #8 with 1525 ELO. The category tests visual understanding plus code generation — a different skill set from text-prompt frontend code where Kimi K3 leads.
OpenAI's First Hardware Is a $230 Keyboard That Shows You What Your Agents Are Doing
Codex Micro is a limited-edition $230 keypad built with Work Louder that lets developers monitor multiple Codex agentic threads at a glance. Translucent top keys cycle through colors to represent each agent's state. It is OpenAI's first branded hardware product.
OpenAI Is Retiring the Entire GPT-4.1 Era: The 2026 Model Graveyard
OpenAI's official deprecation log confirms the full GPT-4.1 family is being wound down in 2026. GPT-4.1-nano fine-tuning windows close in roughly three months. Only GPT-5.x models survive. Developers on gpt-4.1-mini or gpt-4.1 need a migration plan.
GPT-5.6 Sol Pro Closes a 30-Year Gap in Convex Optimization, Lean-Verified in 2.5 Hours
UC Berkeley professor Phillip Kerger used GPT-5.6 Sol Pro to prove that d² function evaluations are the minimum needed for zeroth-order convex optimization, closing a gap open since Protasov's 1996 upper bound. The proof checked out in Lean.