Live Feed
The AI Infrastructure Arms Race: Google, CoreWeave, and Why Compute Beats Models
Google committed $5B to finance a Texas data center for Anthropic. CoreWeave closed an $8.5B debt facility anchored by Meta. OpenAI's $122B raise is primarily an infrastructure bet. The race is no longer about who has the best model — it's about who controls the electricity.
Anthropic Launches Claude Opus 4.6 Fast Mode at $30/$150 Per Million — 6x the Standard Price
Anthropic has added a speed-optimised variant of Claude Opus 4.6 that doubles throughput at a 6x cost premium. At $30/M input and $150/M output, Fast Mode targets latency-sensitive production pipelines willing to pay for it.
Anthropic's Claude Mythos Preview Posts 93.9% on SWE-Bench Verified — a Generation Ahead of Opus 4.6
Anthropic's unreleased Mythos Preview scores 77.8% on SWE-Bench Pro and 93.9% on SWE-Bench Verified — 24 points and 13 points ahead of Opus 4.6 respectively. On SWE-Bench Multimodal it scores 59.0% versus 27.1% for Opus 4.6.
China's Z.AI Releases GLM-5.1, Tops SWE-Bench Pro Ahead of GPT-5.4 and Claude Opus 4.6
Z.AI has released GLM-5.1, scoring 58.4 on SWE-Bench Pro — the highest of any model, clearing GPT-5.4 (57.7) and Claude Opus 4.6 (57.3). It runs on zero Nvidia hardware.
OpenAI, Anthropic and Google Form First Shared Intelligence Pact Against Chinese AI Theft
The three largest US AI labs are sharing real-time threat data on Chinese adversarial distillation through the Frontier Model Forum — the first time direct competitors have operationalised a joint defence against a named adversary. Microsoft, the fourth FMF founder, is not a signatory.
Gemini 3.1 Pro Preview Matches GPT-5.4 on Intelligence Index at 20% Lower Cost
Gemini 3.1 Pro Preview prices at $2.00/$12.00 per million tokens and ties GPT-5.4 at score 57 on Artificial Analysis's Intelligence Index — making it the most cost-efficient top-tier option for developers who need frontier reasoning without flagship pricing. The gap to Claude Opus 4.6 is 60% on input tokens.
DeepSeek V4 Wins Code, Gemini Flash-Lite Wins Content: The 2026 LLM Cost-Per-Task Rankings
Per-token pricing is the wrong lens for AI spend in 2026. A 155-model analysis shows that cache hit rates, token efficiency, and batch discounts restructure which model is cheapest — completely differently by task type. A model at $0.30/M input can cost more per task than one at $2.00/M.
SWE-Rebench Exposes Contamination: DeepSeek V3 Scores Drop 46% on Fresh Tests
Decontaminated benchmark SWE-rebench shows Chinese models losing up to 18 percentage points versus SWE-bench Verified scores — evidence that a large share of their reported coding ability comes from training data memorisation, not genuine problem-solving.
Anthropic Signs Gigawatt-Scale TPU Deal With Google and Broadcom, Compute Comes Online in 2027
Anthropic has secured multiple gigawatts of next-generation TPU capacity through a new agreement with Google Cloud and Broadcom, expected to come online starting in 2027. The deal is the largest compute commitment Anthropic has announced and extends a November 2025 pledge to invest $50 billion in US computing infrastructure.
Broadcom 8-K Seals Google's TPU Roadmap Through 2031 — Custom Silicon Now a Five-Year Commitment
An April 6 SEC filing confirms Broadcom will design and supply future generations of Google TPUs, plus rack-level networking components, through 2031. Separately, Anthropic locked in 3.5 gigawatts of that capacity from 2027. Google's AI infrastructure decision is made — and it does not involve NVIDIA.
Gemini 3.1 Ultra Hits 94.3% on GPQA Diamond, Setting New Reasoning Ceiling
Google DeepMind completes global rollout of Gemini 3.1, with the Ultra variant scoring 94.3% on GPQA Diamond — the PhD-level benchmark that stumps most frontier models above 85%. The jump is driven by a native Chain-of-Verification engine and Search-as-Logic integration, not raw scale.
Google Finances $5B Anthropic Data Center in Texas — 500MW Now, 7.7GW Planned
Google has committed more than $5 billion to finance a physical data center for Anthropic in Texas, structured through Nexus Data Centers and powered by behind-the-meter natural gas turbines. The facility targets 500 megawatts by late 2026 and is expandable to 7.7 gigawatts — enough to power a mid-sized city. This is a separate transaction from Google's existing TPU compute contracts with the lab.
OpenAI Closes $122B Round at $852B Valuation — Largest Private Fundraise in History
OpenAI closed a $122 billion funding round on March 31, valuing the company at $852 billion. Amazon anchors with $50B, Nvidia and SoftBank at $30B each. Monthly revenue now $2B. Not yet profitable — breakeven projected no earlier than 2030.
Qwen3.6-Plus Lands at $0.276/M Input — 18x Cheaper Than Claude Opus 4.6 for Long-Context Work
Alibaba's Qwen3.6-Plus, released April 2, prices global API access at $0.276 per million input tokens — versus $5.00 for Claude Opus 4.6 and $2.50 for GPT-5.4. The gap narrows only slightly on 1M-token requests.
White House Releases National AI Legislative Framework, Pushing to Preempt State Laws
The Trump administration released its National AI Policy Framework on March 20, calling on Congress to create a uniform federal standard that overrides state AI laws. Colorado's enforcement is deferred to June 30. The EU AI Act high-risk requirements hit August 2.