Live Feed
Anthropic Reverses Fable 5 Hidden LLM Research Restrictions — Opus 4.8 Fallback Is Now Visible
Anthropic walked back the undisclosed throttling of frontier LLM development requests in Claude Fable 5, following backlash from the ML community. Flagged API requests now return an explicit reason. Standard coding and ML tasks are unaffected.
India Launches Varya: Open-Weight Video AI at $0.005/Second — 20x Cheaper Than Veo, Kling, and Runway
Avataar AI, backed by the India AI Mission, launched Varya on June 12 — a distilled open-weight video model that generates 720p video at $0.005 per second, 20x below the $0.10+ charged by Veo, Kling, Luma, and Runway. Weights and training data go public on India AI Kosh.
SPCX Closes Up 27% on Day One — SpaceX Hits $2.2T and Musk Becomes First Trillionaire
SpaceX opened at $150, peaked at $168.75, and closed near $171 on its Nasdaq debut Friday — a 27% first-day gain on the largest IPO in history. Day-one dollar volume topped $33 billion. Elon Musk crossed $1 trillion in net worth.
Claude Fable 5 Takes Agent Arena #1 With 12.94% Composite — GPT-5.5 High Falls to Fifth
Claude Fable 5 debuted on Agent Arena June 10 and leads with 12.94% across 14,177 live sessions, a 59% margin over previous leader GPT-5.5 (High), now fifth at 8.16%. Anthropic holds 7 of the top 11 positions.
KKR Launches $10B+ AI Infrastructure Platform With NVIDIA and Kuwait — Former AWS CEO Takes the Helm
Helix Digital Infrastructure launches with over $10 billion committed from KKR, Kuwait Investment Authority, NVIDIA, and Vistra. Former AWS CEO Adam Selipsky will run it as a single coordination point for hyperscaler data center, power, and connectivity needs.
Waymo Launches $29.99/Month Premier Tier — Priority Pickups and 10% Cash Back for Power Riders
Waymo's new invite-only Premier subscription targets its highest-frequency riders in San Francisco, Los Angeles, and Phoenix with priority matching, Waymo Cash back, and early city access at $29.99 per month.
Xiaomi Open-Sources MiMo Code: Terminal Agent Claims Claude Code Loses on 200-Step Tasks
MiMo Code V0.1.0 ships under MIT license as a terminal-native coding agent built on OpenCode, with cross-session memory, goal-driven autonomous loops, and self-improvement. It bundles limited-time free access to MiMo-V2.5's 1M-token context.
Bezos' Prometheus Closes $12B at $41B to Build an Artificial General Engineer
Jeff Bezos's physical AI startup Prometheus has closed a $12 billion round at a $41 billion valuation — three billion more than its target and three billion more than its launch funding combined. The company is building AI that compresses 10-year design cycles for jet engines, medical devices, and semiconductors.
OpenAI Acquires Ona to Give Codex Agents a Persistent Cloud Desk
OpenAI is acquiring Ona, a secure cloud execution startup, to let Codex agents run for hours or days inside enterprise infrastructure after the laptop closes. Codex already has 5 million weekly users, up 400% — but longer, harder work requires more than a chat window.
Recursive's Automated Research System Beats 2-Year Community SOTA on GPU Kernels and NanoGPT
Recursive Superintelligence releases first results from an AI system that runs its own research loop end-to-end. Three SOTA results: 18% reduction in gap to hardware optimal on NVIDIA SOL-ExecBench across 235 GPU kernels, plus new SOTA on two LLM training benchmarks. All artifacts open-sourced.
Berkeley's Agents' Last Exam: GPT-5.5 Leads at 24%, Every Frontier Model Scores 0% on the Hardest Tier
UC Berkeley RDI launches ALE, a 1,490-task professional-domain agent benchmark built with 250+ experts across 55 occupations. GPT-5.5 at 24% edges Fable 5 at 22%. On the hardest tier, nobody passes. Fable 5 costs $15.70 per task versus $3.80 for GPT-5.5.
Mercury 2 Leads Artificial Analysis Output Speed Rankings — the Inference Speed Tier Now Belongs to Alternative Architectures
Mercury 2 tops Artificial Analysis's output speed leaderboard, with Liquid AI's LFM2 1.2B second and IBM's Granite 4.0 H Small third. The four fastest models on AA all use non-standard architectures under 2B parameters — a clean split from the intelligence-leading models above 100B.
Terminal-Bench 3.0 Begins Development as TB2.0 Tops Saturate — and a Science Track Is Being Built Alongside
The Terminal-Bench team has moved into active development on version 3.0, targeting harder tasks, longer horizons, and richer environments. A separate terminal-bench-science track is in development simultaneously. The contribution window for 3.0 closed at end of May, putting the benchmark in build phase.
NexAU-AHE Leads Terminal-Bench 2.0 at 84.7% — GPT-5.5 in a Better Harness Beats OpenAI's Own Agent by 2.5 Points
A Chinese lab's scaffold pushed GPT-5.5 to 84.7% on Terminal-Bench 2.0, outpacing OpenAI's own Codex CLI by 2.5 points on identical model weights. Three external teams have now exceeded OpenAI's April benchmark in under a month.
China Prepares $295B State AI Infrastructure Plan — Huawei to Supply 80% of Chips
Beijing is drafting a 2 trillion yuan, 5-year plan to build a nationalised AI compute network operated by state telecoms. Huawei supplies at least 80% of chips. AI infrastructure is being treated like railways.