Google's Gemini 4 Carbon scored 42 of 43 on our data analytics exam, one more than Gemini 4 Argon. It found every hidden pattern, but it's one of the slowest models we've tested.

Anthropic's Claude Fable 5.2 got all 43 questions right on our data analytics exam, the third model to do so. How it compares with Fable 5.1 and GPT-6.1 Sol on accuracy, cost and speed.

Meta's smallest Muse model scored 38 of 43 on our data analytics exam, tying GPT-6 Luna. Where it did well, where it missed, and why it cost more than its prices suggest.

Which open-weight models can you run on a Mac mini, a DGX Spark, a Mac Studio or a GPU server, and how accurate are they? 22 models from our data analytics benchmark, sorted by the hardware they need.

Five ways Dash 4.5 streaming callbacks and shared storage make data apps better for the people using them, from live collaborative reviews to AI agents that show their work, plus five more ideas.

Plotly Studio Embedded, the AI chat you can switch on in any Dash app on Plotly Cloud, now calls the tools your app exposes with @mcp_enabled. Here’s how to use it.

Inception's Mercury 3, a diffusion language model, is the fastest model we've tested. How it scored on our data analytics exam against GPT-6 Luna and Claude Haiku 5.5.

We tested 12 open-weight models from U.S. labs, from Meta's Muse Spark Open to Thinking Machines' Inkling. How they rank, how they were trained, and how they compare to Chinese open models.

Ricursive's new US open-weight model scored 35 of 43 on our data analytics exam for $2.21. It aced the statistics traps but found none of the hidden patterns.

Tencent's Hy4 scored 42 of 43 on our data analytics exam for $2.11, the cheapest model on our board to reach 42, and joined the cost-accuracy frontier.