172 Billion Tokens Show Every LLM Fabricates Above 10% at 200K Context
The best of 35 open-weight models still invented answers 1.19% of the time about entities that provably did not exist in the source document.…
NVIDIA Nemotron 3.5 Lightning Launches at 1,200 Tokens/Second—30B MoE Outpaces Gemma 4 by 29× and Qwen 3.6 by 35% on Agent Tasks
NVIDIA just shipped a 30B parameter model that outputs 1,200 tokens per second—29 times faster than Gemma 4 26B on identical hardware. This…
Mistral Releases Small 3—24B Open-Source Model Hits 81% MMLU, Runs on Single RTX 4090 at 150 Tokens/Second
A 24-billion parameter model now matches GPT-4o mini while running on hardware you can buy at Best Buy. Mistral just made the $400 billion…
Google’s Gemini 2.5 Pro Hits 82.4% on GPQA Diamond—Beats OpenAI’s GPT-5.5 by 6.1 Points on Graduate-Level Science Reasoning
Google just seized the reasoning crown from OpenAI with a 6.1-point margin on graduate-level science problems. The company that spent 18…
Google DeepMind’s GenCast Beats ECMWF ENS on 97.2% of 1,320 Weather Targets—AI Forecasts 15 Days in 8 Minutes, Outperforms Europe’s Gold Standard by 12.6% on Extremes
Europe’s operational weather forecasting system—the one MeteoSwiss and national meteorological services actually rely on—just got beaten…
Qualcomm in Talks to Acquire Tenstorrent for $8-10 Billion—Jim Keller’s RISC-V AI Chip Startup Valuation Triples in One Year
Qualcomm just offered 3x last year’s valuation for a RISC-V chip startup. The bid tells you more about NVIDIA’s moat than…
Google DeepMind Launches AlphaGenome on June 25—AI Processes 1 Million DNA Base Pairs at Single-Nucleotide Resolution, Wins 22 of 24 Benchmarks
DeepMind just made every specialized genomics model obsolete in a single release. AlphaGenome processes 1 million DNA base pairs at once—and…
Anthropic Raises $65 Billion at $965 Billion Valuation—Run-Rate Revenue Hits $47 Billion as Series H Becomes Largest AI Funding Round
Anthropic just raised more in a single funding round than the entire GDP of Croatia. At $965 billion, they’re worth more than every bank…
DeepSeek AI Releases Manifold-Constrained Hyper-Connections Architecture—New Transformer Design Cuts Overfitting and Computational Cost Through Dynamic Neuron Links Within Mathematical Manifolds
A Beijing lab just made every transformer architecture from 2017-2025 look like a brute-force approximation. Two competing labs shipped…
Meta Llama 4 Launches April 5 with 109B-Parameter Scout Model—17B Active Parameters Fit on Single H100, Outperform GPT-4.5 and Claude 3.7 on STEM Benchmarks
Meta just made frontier AI fit on one GPU—and simultaneously banned 450 million Europeans from using it.