ModelsLing 3.1 Flash Guide: A Point Release With Five Times the Compute
Ling 3.1 Flash keeps the name of a cheap model but grows to 560B total and 25B active parameters. What the extra size buys, what it costs, and when 3.0 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.
Models
ModelsLlama 3 8B Lunaris: What a Four-Model Merge Is Good For
Lunaris is an 8B Llama 3 merge for chat and roleplay at a few cents per million tokens. How it was built, why 8K context is the real limit, and where it fits.
Read
ModelsLlama 3.1 70B Instruct in 2026: When an Old Model Still Fits
Llama 3.1 70B shipped in July 2024 and Llama 3.3 70B beat it at the same size. Why it is still served, what its licence asks of you, and when to pick it today.
ReadModelsLlama 3.1 8B Instruct: The Small Default That Will Not Retire
Meta released Llama 3.1 8B in July 2024 and it is still a common base for cheap tasks and fine-tunes. What it does well, where it breaks, and when to move up.
ReadModelsLlama 3.1 Euryale 70B v2.2: A Roleplay Fine-Tune, Explained
Sao10K's Euryale v2.2 is a Llama 3.1 70B tuned for roleplay and creative writing. What changed from v2.1, the licence catch, and the sampler settings it expects.
ReadModelsLlama 3.2 1B Instruct: What a One-Billion Model Is Actually For
Llama 3.2 1B was pruned and distilled from Llama 3.1 8B to run on phones. What it can do, what it cannot, and why calling it over an API is usually the wrong test.
ReadModelsLlama 3.2 3B Instruct: The Smallest Llama That Can Call Tools
Llama 3.2 3B is the point where a small Meta model starts handling function calls and multilingual tasks. Specs, measured trade-offs, and how it compares to 1B and 8B.
Read
ModelsLlama 3.3 70B Instruct: A Post-Training Upgrade Worth Knowing
Llama 3.3 70B kept the 3.1 architecture and improved the post-training. Where it gained, where it stood still, and how it compares with Llama 4 Scout.
Read
ModelsLlama 3.3 Euryale 70B: The Successor and Its Trade-offs
Euryale v2.3 moves Sao10K's roleplay line onto Llama 3.3 70B Instruct. How it was trained, what the 16K training length means, and what to own before you deploy it.
ReadModelsLlama 4 Maverick: 400B Total, 17B Active, and a Leaderboard Lesson
Meta's Llama 4 Maverick is a 128-expert multimodal MoE with 1M context. Its architecture, its real benchmark profile, the LMArena episode, and the licence terms.
Read
ModelsLlama 4 Scout: What the 10M Token Context Means in Practice
Llama 4 Scout advertises a 10M-token window, but it was trained at 256K and most hosts serve far less. Specs, hardware fit, and how to test long context yourself.
ReadModelsLlama Guard 4 12B: A Multimodal Safety Classifier, Read Honestly
Llama Guard 4 is a 12B classifier pruned from Llama 4 Scout that labels prompts and replies safe or unsafe. What it catches, what it misses and how to wire it in.
ReadShowing 169–180 of 731 articles