↓ Skip to main content

Models

featured.jpg

3 min

Kimi K3.1 Pro weights are out. Should you swap it into your stack?

Moonshot AI published Kimi K3.1 Pro weights on Hugging Face. Here is who should care, what changes, and the one test to run before you swap …

featured.jpg

3 min

Can Your Model Keep State? YANAMI Is a New Multiturn Multimodal Benchmark

YANAMI is a new multimodal dataset and benchmark for multiturn, stateful interaction. Here is what that tests and why single-prompt evals …

featured.jpg

3 min

Mistral Large 4 Is Getting Open Weights. Can You Ship On It?

Mistral says it is releasing the weights of Mistral Large 4. Weights are a download, not a deployment. Here is what stands between the file …

featured.jpg

3 min

What Happens When RSI-Opt's Agents Improve How Models Train?

RSI-Opt points self-improving agents at model training itself. Here is the three-step trend behind it and the one thing that decides whether …

featured.jpg

3 min

Why Grok Picks Republicans and OpenAI Picks Democrats When Forced

TechCrunch reports that Grok picks Republicans and OpenAI's model picks Democrats when forced to choose. Here is why that happens and how to …

featured.jpg

2 min

ModelCompare: Why Every AI Model Gives the Same 'Controversial' Take on Humans

ModelCompare says every AI model gives the same 'controversial' take on humans. The cause is a shared training recipe, and it breaks …

featured.jpg

3 min

OpenAI Models Are Coming to Google Cloud, and 'Coming' Is the Operative Word

OpenAI says its models are coming to Google Cloud. The verb is future tense, so here is what would have to be true before you change …

featured.jpg

2 min

Why Is the RAP Poster Dataset Built on Markup, Not Pixels?

RAP, a paper-to-poster dataset, targets declarative markup instead of images. Here is why that design choice matters for anyone generating …

featured.jpg

3 min

Build a Scam-Message Classifier in One Sitting, Inspired by DeepMind's Singapore Work

DeepMind and Singapore are building a fine-tuned scam-detection model. Here is a prompt-first baseline classifier you can build and test …

featured.jpg

3 min

Gemini 4 Argon Has a Name and a Page. Where Are the Numbers?

Gemini 4 Argon is listed on DeepMind's Gemini page, but nothing in the source gives a benchmark, price or limit. Here is what to do before …

featured.jpg

3 min

Karpathy Says the LLM Capability Gap Is Widening. Who's on Which Side?

Karpathy says the gap in shared understanding of LLM capability is widening. Here is why it happens and how to check which side of it your …

featured.jpg

3 min

Build your own decision model: replace an LLM call this afternoon

A Hacker News post titled 'Build your own decision model' is a prompt to stop paying an LLM for routing calls. Here is a recipe and the code …

featured.jpg

3 min

Plan-and-Patch Asks Whether Diffusion Language Models Should Write Agent Plans

Plan-and-Patch pitches diffusion language models as agent planners that can fix one bad step instead of regenerating the whole plan. Here is …

featured.jpg

3 min

Should You Switch to Mistral Large 4 This Week?

Mistral published Large 4. Before you repoint production traffic, here is what the release history says about how to handle a new flagship.

featured.jpg

3 min

Can Anthropic Report Your Claude Diary Entry to Police? One Woman Faces a Felony

TechSpot reports Anthropic flagged a Claude diary entry to police and a woman now faces a felony charge. Here is what that confirms and what …

featured.jpg

2 min

Why Stratego Stumped AI Until Now, and Why the Budget Matters

Ars Technica reports AI has beaten the best Stratego player in history on a budget. Here is why hidden information was the wall, and who …

featured.jpg

3 min

Can You Build on Google's Gemini 4 Argon Yet? No

Google announced Gemini 4 Argon, but Ars Technica reports you can't use it yet. Here is what that means for anyone planning a migration, and …

featured.jpg

3 min

Jeff's 0.8B decision models claim ~30 ms, trained at home

Jeff, a GitHub repo of 0.8B decision models trained at home, claims ~30 ms. Here is the trend it belongs to and what to test before trusting …

featured.jpg

2 min

How To Turn On Gemini 3.8's New Live Avatar Face Today

Google's Gemini 3.8 Live now shows an animated Live Avatar face during voice chats — here's how to switch it on and test it.

featured.jpg

2 min

Google's Gemini 3.8 Live Gets A Face, And Builders Get New Work

Google added a Live Avatar to Gemini's real-time Live API, turning voice-only interfaces into a design decision developers now have to make.

featured.jpg

3 min

Ovis-Embedding Claims Universal Omni-Modal Search. Does It?

A new paper claims one embedding space for text, image, video and audio. Here's the mechanism that makes 'universal' hard to prove.

10 min

The Maturity Curve: How AI Models Stopped Competing on Intelligence

Five hundred and forty-five open-weight models, an Elo spread of two hundred points at the top, and a price spread of nearly ninety times. …

featured.jpg

3 min

A 4B Model Just Beat Postgres's Query Planner by 81%

A 4B model reportedly plans Postgres queries 81% faster than Postgres's own optimizer — as a benchmark, not a shipped feature.

featured.jpg

3 min

ZGCM-1 Claims Full Openness. Here's What That Actually Requires

ZGCM-1's arXiv title claims 'fully open and extremely efficient.' Here's what's actually verified so far, and what would close that gap.

featured.jpg

2 min

Why Insight Partners Is Betting Against the OpenAI Consensus

Insight Partners' Deven Parekh explains why the firm is spreading its AI bets instead of concentrating on OpenAI and Anthropic like everyone …

featured.jpg

2 min

Why Real-SWE Tests Coding Models On Your Own Codebase

Real-SWE benchmarks AI coding models on private enterprise repos, not public GitHub — here's how to build your own version this week.

featured.jpg

2 min

Garry Tan's Distillation Call Admits US Open-Weight Labs Are Behind

Garry Tan's call for US open-weight labs to distill frontier models reveals how far behind they already are.

featured.jpg

2 min

Can Mathematicians Still Trust OpenAI With Unpublished Proofs?

A Mathstodon post revives the FrontierMath funding controversy, showing mathematicians keep learning OpenAI's benchmark terms only after …

featured.jpg

2 min

Mistral Raises €3 Billion To Be Sovereign And Frontier At Once

Mistral's €3B raise pairs sovereign, open-weight, and frontier claims in one slug — a combination no AI lab has held together yet.

featured.jpg

3 min

A New arXiv Paper Says Quantization Quietly Breaks Model Memory

A new arXiv paper names a real quantization failure mode in recurrent-state memory but doesn't publish accuracy numbers to size the damage …