Blog

My hands-on StepFun Step 5 Preview review: a 600B/27B MoE reasoning model at $1/$2.70 per 1M tokens. The pricing is real, the benchmarks are self-graded. Here is my verdict.

StepFun Step 5 Preview is a 600B/27B sparse MoE model with a 1M-token context, open weights due Oct 15. Here are the specs, pricing, benchmarks, and my take.

Qwen Image 2.1 is a great open model with a license catch. Here are the 8 best Qwen Image 2.1 alternatives in 2026, with real API prices and who each is for.

A hands-on Qwen Image 2.1 review: what the 7B open-weight model does well, the native transparency trick, and the license catch nobody's mentioning.

Qwen Image 2.1 is Alibaba's new 7B open-weight image model with native transparency. Here's what changed, how to run it, the cost, and the license catch.

Laya AI is a fast open-source decision model, but it needs fine-tuning and only classifies. Here are 8 Laya AI alternatives, from managed APIs to teammates that act.

Laya AI is Convai's open-source, 421M System 1 decision model: 33ms typed answers with calibrated confidence, no text generation, and an honest zero-shot caveat.

TypeSafe calls Jev an ultrafast System One model at 70-500ms a decision. Here is what the speed claim really means, where it holds up, and where it does not.

Claude Code now reads AGENTS.md natively. Here's how AGENTS.md and CLAUDE.md relate, which file wins, and how to write one that agents actually follow.