Live Feed
DeepSeek Confirms Significant API Price Hike — V4-Flash Demand Overwhelmed Its Proof-of-Concept Pricing
DeepSeek's V4-Flash 0731 became Ollama's fastest-growing model ever at $0.14/$0.28 per million tokens, then the company announced prices would rise significantly. At current rates it is 105x cheaper per task than Claude Fable 5 — but Meta Muse Spark and GPT-5.6 Luna now compete on capability and price.
Microsoft FY2026 Filings Reveal OpenAI Drove $24.1B — Roughly 70% of Its Actual AI Revenue
A first-time disclosure in Microsoft's FY2026 regulatory filings shows OpenAI commercial arrangements generated $24.1 billion in revenue — about 70% of Microsoft's actual AI sales, per Bloomberg Intelligence. The company has also invested $11.9 billion into OpenAI.
Finance Agent Harness Lifts GPT-5.5 From 55.8% to 79.1% on BigFinanceBench — Retrieval Beats Model Upgrades
Primer's domain-specific agent wrapping GPT-5.5 scored 79.1% on 928-question BigFinanceBench versus 55.8% for the raw model — a 23.3-point lift that newer frontier models have not closed. Sixteen of those points came from retrieval alone.
Arista Revenues Surge 38% as AI Data Center Ethernet Demand Outpaces GPU Supply
Arista Networks posted a 38% revenue jump driven by AI data center switch demand from Meta, Microsoft, Oracle Cloud, Anthropic, and Google. Its Etherlink and 7700/7800 series now sit inside the biggest frontier AI clusters in production.
Volta Closes $100B AI Factory Deal in Norway at 7 Months Old — 133MW, Vera Rubin, Mystery Lab
NVIDIA-backed Volta signed a $100B strategic cooperation with an unnamed AI lab and Bitdeer to deploy 133MW of Vera Rubin compute in Norway. Anthropic declined to comment. The fastest-capitalized compute structure play of 2026.
Artificial Analysis Endpoint Accuracy Index: Same Open-Weight Model, Different Scores by Provider
Artificial Analysis launches a new benchmark measuring how much serverless providers preserve open-weight model accuracy after quantization and custom kernel tuning. Coverage opens with GLM-5.2, gpt-oss-120b, and DeepSeek V4 Pro. Reference parity is 100%.
Anthropic Posts Jobs for a Custom Silicon Team — Claude's Hardware Stack Is Going In-House
Anthropic is actively hiring chip designers for a dedicated custom silicon team, moving beyond contracted compute from AWS, Google, Nvidia, and AMD toward co-designing hardware and models together.
Nashville Votes to Use Eminent Domain Against a Data Center — Community Opposition Finds a New Legal Tool
Nashville's Metro Council approved eminent domain acquisition of properties near the Nashville Zoo to block a proposed DC BLOX data center. It is the first US city to deploy the tool offensively against an AI infrastructure project.
Qwen 3.8 Max Leads LiveBench Agentic Coding at 64.6 — Beats Fable 5 Max Effort at $0.28 Per Task
Alibaba's Qwen 3.8 Max scores 78.5 overall on LiveBench, placing #6 globally, but its standout number is 64.6 on Agentic Coding — above Claude Fable 5 Max Effort and Kimi K3 at 62.2. Cost per successful task is $0.275.
Meta Enters Coding Agents With Muse Code and Muse Spark 1.2 — Async Subagents, Replay-Safe Runtime
Meta launches Muse Code, a terminal coding agent powered by the new Muse Spark 1.2 model. The agent runs persistent background subagents throughout each session, not just for individual tasks, and uses a local event log for crash-safe resumption. Muse Spark 1.2 scores 54 on the AA Intelligence Index at $1.25/$4.25 per million tokens.
Cloudflare OS: Open-Source Agent Platform Targets the Enterprise Orchestration Layer
Cloudflare open-sourced Cloudflare OS, a governance and orchestration platform for AI agents positioned as the missing infrastructure between agents and enterprise systems of record. It ships with Gatekeepers for access control, MCP Server Portals for existing MCP integrations, and a zero-trust governance layer. Managed deployment follows.
DeepMind's Brain Drain: Hassabis Steps Back, Jeff Dean Leaves to Found a Startup
Alphabet CEO Sundar Pichai confirmed sweeping DeepMind leadership changes: Demis Hassabis becomes Chair and Alphabet Chief Scientist, Jeff Dean departs to co-found Discovery Loop alongside Oriol Vinyals, Quoc Le, and Sanjay Ghemawat. Koray Kavukcuoglu takes over DeepMind operations and will oversee Gemini 4.
OpenAI Calls Apple's Lawsuit 'Careless, Aggressive and Oddly Personal' — Publishes Raw Emails
OpenAI published the internal messages behind Apple's trade-secret suit in a blog post titled 'Apple Is Getting This Wrong,' calling the lawsuit 'careless, aggressive and oddly personal.' OpenAI says Apple emailed the wrong person before filing, that allegations about Tang Tan and Chang Liu are factually wrong, and that Apple had 5 months of silence before the suit landed.
GPT-5.6 Sol Rewrote Its Own Production Kernels — 20% Inference Cost Cut, 15% Throughput Gain
OpenAI's post-launch efficiency blog details how GPT-5.6 Sol used Codex to autonomously rewrite its own GPU kernels in Triton, redesign speculative decoding, and tune load balancing. Kernel work alone cut serving costs 20%. Self-trained speculative decoding added 15% throughput. The same model optimized the stack it runs on.
Google's $43.8B Guarantee Network Gives TPU Operators a 2.2-Point Rate Edge Over Nvidia
An FT investigation reveals the $200B off-balance-sheet financing structure under Google's TPU strategy: Alphabet backstops data center leases, cutting borrowing costs to 7.1% for TPU operators versus 9.3% for Nvidia-ecosystem projects. At $35B scale, the gap is $770M per year in extra interest — compounding as the buildout scales.