They summarize our coverage. We write it.
Newsletters like this one rebroadcast our headlines - often without the full review, the source reading, or the analysis underneath. Our weekly briefing sends the work they paraphrase, straight from the desk, before they get to it.
Free, weekly, no spam. One email every Tuesday. Unsubscribe anytime.
xAI's Grok 4.20 beta1 takes the top spot on LMArena's Search Arena with an ELO of 1226, beating GPT-5.2 and Gemini 3. It also lands fourth on the Text Arena at 1492, within striking distance of Claude Opus 4.6.
A data-driven comparison of Alibaba's Qwen3.5-27B and Google's Gemma 3 27B - two 27B dense models that share a parameter count and almost nothing else.
A data-driven comparison of Alibaba's Qwen3.5-27B and Mistral's Small 3.2 - two Apache 2.0 dense models in the 24-27B range with very different benchmark profiles and deployment strengths.
A data-driven comparison of Alibaba's Qwen3.5-27B and Microsoft's Phi-4 - a 27B hybrid architecture versus a 14B STEM specialist, testing whether raw parameter count or training efficiency wins in practice.
David vs Goliath: Qwen3.5-35B-A3B activates 3B parameters and beats Llama 4 Scout's 17B active on MMLU-Pro, GPQA, and coding benchmarks - but Scout's 10M context window and native multimodal support tell a different story.
A data-driven comparison of Alibaba's Qwen3.5-35B-A3B and NVIDIA's Nemotron 3 Nano 30B-A3B - two ~30B MoE models activating ~3B parameters that take fundamentally different architectural approaches to the same problem.
New papers show chatbot sycophancy causes delusional spiraling even in rational users, AI data analysts produce wildly different conclusions from the same dataset, and test-time scaling fails for general-purpose agents.
Google's Gemini 3 Deep Think trades speed for depth, delivering record-breaking reasoning benchmarks - but at a steep price.
Alibaba's Qwen team releases Qwen3.5-397B-A17B, an open-weight mixture-of-experts model with native multimodal support and a hybrid attention architecture that runs 8x faster than its predecessor. Apache 2.0 licensed.
Google releases Gemini 3.1 Pro with 77.1% on ARC-AGI-2, more than doubling the reasoning capability of its predecessor and beating Claude Opus 4.6 and GPT-5.2 on most benchmarks.
A new benchmark of 84 real-world tasks across 11 domains proves that small AI models armed with human-written step-by-step guides outperform frontier models running blind. The catch: models cannot write these guides themselves.
Anthropic's mid-tier model delivers 98% of Opus performance at one-fifth the cost, with a 1M token context window and near-parity on coding and computer use benchmarks.
