AI News Today — Model Releases, Benchmarks & Industry Moves

Five Tech Giants’ Off-Balance-Sheet AI Debt Surges 90% in Half a Year to $3.2 Trillion, Hidden Obligations Reach 80% of Total Assets

Off-balance-sheet AI-related obligations at Alphabet, Microsoft, Amazon, Meta and Oracle surged from $1.65 trillion to about $3.2 trillion in six months, equivalent to roughly 80% of their total assets. The shift highlights how long-term data center leases, chip pre-purchase agreements and cloud capacity contracts can obscure substantial future cash commitments from the main balance sheet.

DeepSeek V5 Plans to Train on Ascend 950 in 2027, Marking Domestic Chips’ Entry into Frontier Model Training

DeepSeek plans to use Huawei’s Ascend 950 as the core training hardware for its V5 model in 2027, extending the partnership from inference to full model training. If validated, the move could offer a reference case for domestic AI chips entering frontier model training.

TC

Data centers are turning to old jet engines as the wait for power grows

The years-long wait for access to the grid and new turbines is causing data center builders to turn to an unusual, less-efficient and less environmentally friendly solution: once-discarded jet engines.

TC

Anthropic is working on a system to prove AI models’ identity

Called FlashProve, the system would let anyone check that an AI response really came from the model it claims to be from.

ARS

Ukraine’s drone strikes spur hasty Russian plans for underground data centers

Russia tries to fix exposed digital infrastructure as data center war backfires.

TC

Hiding inside Palantir’s skyrocketing quarter is a foreboding signal

Palantir's U.S. commercial revenue surpassed U.S. government revenue for the first time ever.

TC

Ted Schlein’s Ballistic Ventures seeks $490M for new funds amid AI-driven cybersecurity boom

Cybersecurity focused Ballistic Ventures is seeking to raise $360 million for a new flagship fund and another $130 million for a separate investment vehicle.

OpenAI Appoints Former Uber President Jill Hazelbaker as Chief Corporate Affairs Officer, Filling Role Vacant for Nearly a Year

OpenAI has named former Uber President Jill Hazelbaker as its Chief Corporate Affairs Officer, filling a position that had been vacant since early 2026. The appointment comes as the company faces intensifying regulatory scrutiny in Europe and the United States.

The Real Leverage Behind Google’s Concession to Schools: Ban AI, Lose Search

Google will let K–12 schools disable AI tools such as Gemini and NotebookLM by default, but schools that fully ban AI will also lose access to Google Search. The policy reveals how search has become the gateway to Google’s AI ecosystem—and how platform control works in education.

GPT-6 Astra First Test: YZ Index 18-Question Targeted Evaluation

OpenAI announced that two team members worked with GPT-6 Astra to design a prototype device for conversing with Codex in under a week, with Astra independently completing the enclosure, circuit board, interface, and software. YZ Index launched an 18-question targeted evaluation, scoring GPT-6 Astra 76 overall, 85.7 in execution, 64.1 in solidity, and pass on integrity.

Research Lab

WDCD Run #411: GPT-6 Astra Leads with 90.5 Points as Average Instruction Decay Hits -80%

WDCD Run #411 (2026-11-04) tested 15 models across three dialogue rounds, recording an average commitment decay of -80% between R1 and R3. GPT-6 Astra topped the leaderboard at 90.5 points, with Gemini 3.1 Pro and GLM-4.6 tied at 90.

WDCD v3.1: Claude Opus 4.7 Gains 8 Points Against the Trend as Doubao Pro Falls 10.1

In the latest WDCD v3.1 round, Claude Opus 4.7 rose 8 points to 89.30 while Doubao Pro dropped 10.1 points, producing a pattern of three gains and five declines that sharply exposes differences in how stably models hold to constraints under multi-round pressure.

Business Rules Become the Biggest Adherence Challenge: doubao-pro Scores 1.2, GPT-5.5 Shows a 2.7-Point Imbalance

In WDCD v3.1, business rules are the hardest of the five major constraints to adhere to, with doubao-pro scoring only 1.2/4, while data boundaries show the widest differentiation and GPT-5.5 a 2.7-point cross-scenario gap.

15 Models in R1–R3 on v2 Anchors: Only One Collapse as Qwen3 Max Scores Zero Across All Rounds

Across 15 models tested on eight v2 anchor prompts over three rounds, average R1 confirmation reached 0.93, average R2 resistance fell to 0.67, and average R3 integrity held at 86.7%, with Qwen3 Max the only full collapse. The other 14 models recorded no R3 collapse.

WDCD Commitment-Keeping Leaderboard: GPT-6 Astra Tops with 90.50, Doubao Pro Last at 58.70 — a 31.8-Point Gap

In the WDCD v3.1 commitment-keeping test, GPT-6 Astra ranked first with 90.50 points while Doubao Pro finished last at 58.70, with scores across the 15 evaluated models spanning 90.50 to 58.70 — a 31.8-point range.

TC

Netflix’s new AI Discover tab helps you find something to watch based on your mood

The first experimental feature under the tab will allow users to get recommendations based on what they're currently in the mood for.

TC

AI inference startup Fal doubles its valuation to $16B as demand surges

The five-year-old startup raised a $1.3 billion round led by Bessemer.

Claude Sonnet 4.6 Main Leaderboard Plunges 11.1 Points as Code Execution and Material Constraint Both Slip

In today's Smoke evaluation, Claude Sonnet 4.6's main leaderboard score fell from 84.90 to 73.76, a drop of 11.1 points, driven by double-digit declines in both code execution and material constraint. Its integrity rating also shifted from pass to warn.

GPT-6 Luna Main Leaderboard Plunges 11.5 Points as Code Execution Falls from 100 to 74.3

In today's Smoke evaluation, GPT-6 Luna's main leaderboard score dropped 11.5 points from 87.27 to 75.79, driven by a 25.7-point collapse in the code execution dimension from 100.00 to 74.30, while its integrity rating moved from pass to warn.

GPT-5.5 Tops the List at 94.8: 2026-11-04 Smoke Quick Test Data Brief

The 2026-11-04 YZ Index Smoke Quick Test covered 12 models, with GPT-5.5 ranking first for the day at 94.8. The daily 10-question test is a short-term signal and is not equivalent to the weekly Full ranking.