Live Feed
Anthropic Draws $800B Investor Offers as OpenAI Disputes the Revenue Behind the Number
Investor offers have valued Anthropic at up to $800 billion — more than double its February raise — while OpenAI's chief revenue officer circulated an internal memo accusing Anthropic of inflating its $30 billion run rate by $8 billion through gross accounting practices.
100 Teams, 300 Humanoids: Beijing's Robotic Half-Marathon Goes Autonomous on April 19
The second annual Beijing E-Town Humanoid Robot Half-Marathon runs April 19 with 100+ competing teams — five times last year's field — and nearly 40% of entrants operating under full autonomous navigation. Remote-controlled robots face a 1.2× time penalty designed to make human supervision economically uncompetitive.
Arena Opens Image-to-WebDev Leaderboard; Kimi K2.5 and Grok 4.20 Beta Enter Document Rankings
Chatbot Arena launched its Image-to-WebDev evaluation on April 15, testing models on visual-to-code conversion. The Document Arena simultaneously added Moonshot's multimodal Kimi K2.5 and xAI's Grok 4.20 Beta — expanding the most credible comparative view of how frontier models handle long-form document tasks.
AI Coding Agents Ship Insecure Code 87% of the Time — Even When All Tests Pass
Endor Labs measured 200 real tasks across 108 open-source projects and found that 87% of AI-generated code contains at least one exploitable vulnerability, regardless of whether functional tests pass. The best security performer, OpenAI Codex with GPT-5.4, scored just 17.3% on security correctness.
Fluidstack Raises $1B at $18B Valuation — AI Data Center Demand Triples Its Value in Four Months
The specialized AI data center startup jumped from a $7.5B valuation in December 2025 to $18B, making it one of the fastest-appreciating infrastructure bets in tech. The $1B round reflects mounting hyperscaler capacity constraints with no relief in sight.
Google Ships Gemini 3.1 Flash TTS: Audio Tags, 70+ Languages, 1,211 Elo on AA Leaderboard
Google DeepMind has launched Gemini 3.1 Flash TTS in preview, bringing director-level voice control through inline audio tags, native multi-speaker dialogue, and 70+ language coverage. It ranks second on the Artificial Analysis TTS leaderboard at Elo 1,211 — beating ElevenLabs v3 — at $1.00/M text tokens and $20.00/M audio tokens.
Google Launches Native Gemini App for macOS — The Last Major AI Lab on Desktop
Google has shipped a 100% native Swift Gemini app for macOS 15+, with an Option+Space hotkey, live screen sharing, and access to Gemini 3 Fast, Thinking, and Pro models. ChatGPT and Claude Mac apps arrived roughly a year ago.
Jane Street Commits $7B to CoreWeave — Wall Street's Biggest Quant Shop Is Now a Frontier Lab Client
Jane Street signs a $6B AI cloud agreement with CoreWeave and takes a $1B equity stake at $109 per share, making the quantitative trading firm the fifth-largest shareholder in the AI cloud provider. The deal includes access to next-gen NVIDIA Vera Rubin compute and marks CoreWeave's third major contract announcement in seven days.
OpenAI Launches GPT-5.4-Cyber With Thousands of Vetted Defenders — Directly Contrasting Anthropic's 40-Partner Glasswing
GPT-5.4-Cyber, a fine-tuned variant of OpenAI's flagship model with reduced refusal boundaries for security work, launches one week after Anthropic's Mythos announcement. BNY, Citi, CrowdStrike, NVIDIA, Oracle, Cisco, and Zscaler are among day-one partners. $10M cybersecurity grant program attached.
TeraWulf Raises $900M to Convert Bitcoin Mining Infrastructure to AI Data Centers
TeraWulf priced approximately $900 million in new equity on April 15 to fund an AI/HPC campus in Houseville, Kentucky — the latest Bitcoin miner betting its power contracts and cooling infrastructure are worth more as AI compute than as crypto.
GPT-5.4 and Claude Opus 4.6 Tie at 81.8% on Terminal-Bench 2.0 — Scaffold Engineering Now Determines the Winner
With identical ForgeCode scaffolding, GPT-5.4 and Claude Opus 4.6 score exactly 81.8% on Terminal-Bench 2.0. Gemini 3.1 Pro trails 1.6 points behind via TongAgents. At the frontier, the agent wrapper has become as decisive as the model.
Claude Agents Outperform Anthropic's Own Alignment Researchers — 97% vs 23% at $22/Hour
Nine parallel Claude Opus 4.6 agents closed 97% of an alignment research performance gap in five days for $18K total. Two human researchers spent seven days on the same problem and reached 23%. The result is the first credible evidence that automated alignment research is practical on measurable problems.
EU AI Act Full Enforcement Is 16 Weeks Away — and Most Companies Are Already Behind
August 2, 2026 is the general application date for the EU AI Act's high-risk requirements. But the Act's most common planning error is treating that date as a starting line. Key obligations have been live since February 2025, and companies that haven't moved are non-compliant now.
Gemini API vs GPT-5.4: The April 2026 Developer Economics
Google's Gemini API is 20-40% cheaper than OpenAI across comparable tiers, has a more generous free tier, and leads on long-context processing. GPT-5.4 still wins on coding benchmarks and has the larger third-party ecosystem. The data as of April 2026.
DeepMind Ships Gemini Robotics-ER 1.6 With Instrument Reading for Industrial AI
Google DeepMind's updated embodied reasoning model adds gauge and sight-glass reading — developed in partnership with Boston Dynamics — plus multi-view task success detection and spatial pointing. Available now via the Gemini API.