ModelsKAT-Coder-Pro V2.5 Guide: Kwaipilot's Closed Agentic Coder
Kuaishou's coding model claims 65.2 on SWE-Bench Pro at a fraction of frontier prices. What the claim covers, how it was trained, and when the open Dev sibling is enough.
ReadNotes on models, agents, and what inference costs
Practical writing for developers building with large language models — how they work, how to pick one, and how to keep the bill predictable.
Picking between models — and testing them on your own work.
Models
ModelsKimi K2 Guide: The 0711 and 0905 Snapshots in 2026
Kimi K2 started Moonshot's trillion-parameter line in July 2025. What the 0711 and 0905 snapshots differ on, what the licence asks, and whether K2 still earns a slot.
Read
ModelsKimi K2 Thinking Guide: Long Tool Chains and Native INT4
Kimi K2 Thinking added long-horizon reasoning to K2 and claimed 200 to 300 sequential tool calls. What it does well, what its INT4 weights change, and what has replaced it.
Read
ModelsKimi K2.5 Guide: Native Vision, Agent Swarm and Two Modes
Kimi K2.5 gave Moonshot's K2 line native vision and an experimental agent swarm. What changed from K2 Thinking, what the swarm claims mean, and when K2.6 is the better call.
ReadModelsKimi K2.7 Code Guide: Fewer Thinking Tokens, Same Price
Kimi K2.7 Code is a coding-only cut of Moonshot's K2 line that thinks about 30% less than K2.6. How to read its benchmarks, and when to choose it over K2.6 or K3.
ReadModelsLaguna S 2.1 Guide: Poolside's 8B-Active Agentic Coder
Poolside's 118B-A8B open-weight coding model with a 1M window. What it scores, what its own launch post admits, and when to pick it over Laguna XS 2.1.
ReadModelsLaguna XS 2.1 Guide: A 3B-Active Coding Agent for Your Desk
Poolside's 33B-A3B open-weight coding model, what improved over XS.2, how it compares with Qwen3.6 35B-A3B, and the setup details that trip up local runs.
Read
ModelsLFM2.5-2.6B: Liquid AI's Small Reasoning Model for Agents
A 2.6B hybrid convolution model that always reasons before answering and calls tools in Python syntax. What it does well, where it is thin, and how to run it.
ReadModelsLing 3.0 Flash Fin Guide: A Finance Model You Should Test First
Ant Group tuned Ling 3.0 Flash on financial data for research, filings and spreadsheet work. What the Fin variant claims, what it does not show, and how to test it.
ReadModelsLing 3.0 Flash Guide: 5.1B Active Parameters, MIT Weights
inclusionAI's Ling 3.0 Flash is a 124B MoE that activates about 5.1B parameters per token. What the hybrid attention buys, what it costs, and when to use it.
ReadModelsLing 3.0 Flash Sante Guide: A Free Medical Model, Unbenchmarked
Ling 3.0 Flash Sante is inclusionAI's health-tuned MoE, launched with a free endpoint. What is known, what is missing, and the medical jobs it is and is not safe for.
Read
ModelsLing 3.0 Flash VL Guide: Image and Video Input on 5.5B Active
Ling 3.0 Flash VL adds image and video input to a 124B MoE for about 0.4B extra active parameters. How it is built, the context caveat, and when to pick it.
ReadShowing 157–168 of 446 articles