ModelsQwen2.5 VL 72B Instruct: Documents, Charts and Grounding
Qwen's January 2025 vision-language flagship reads documents, charts and screens well. Its benchmarks, its resolution cost dial, and the licence that is not Apache.
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
ModelsQwen3 14B Guide: One Dense Model, Two Ways to Answer
Qwen3 14B is a 14.8B dense, Apache 2.0 model that can think before answering or reply directly. How the switch works, what it costs, and when to pick 8B or 32B.
Read
ModelsQwen3 235B A22B Guide: The Original and the 2507 Instruct
Qwen3's flagship MoE shipped as a hybrid thinker in April 2025, then split. What 235B total and 22B active mean, what 2507 changed, and which ID you should call.
Read
ModelsQwen3 235B A22B Thinking 2507: Reasoning Without a Switch
The July 2025 thinking-only Qwen3 flagship: where its reasoning scores land, why output budget is the real cost, and the parsing quirk that trips integrations.
Read
ModelsQwen3 30B A3B Guide: The Hybrid MoE That Started a Line
The original April 2025 Qwen3 30B A3B: one set of weights, two modes, 3.3B active parameters. What it is, how the switch works, and when it still beats its 2507 heirs.
Read
ModelsQwen3 30B A3B Instruct 2507: The Non-Thinking Rebuild
Qwen dropped the hybrid switch and rebuilt its 30B MoE as a model that never thinks. What the 2507 update changed, its 256K context, and when to pick it over Thinking.
Read
ModelsQwen3 30B A3B Thinking 2507: Reasoning at 3.3B Active
A reasoning-only MoE that activates 3.3B parameters per token and thinks for longer than its predecessor. Where the extra tokens pay, what they cost, and how to parse them.
ReadModelsQwen3 32B Guide: The Dense Flagship of the 2025 Lineup
Qwen3 32B is the largest dense model in the April 2025 Qwen3 release. How it differs from the 30B A3B MoE, what it costs to host, and why dense still suits fine-tuning.
ReadModelsQwen3 8B Guide: A Small Hybrid Model You Can Run Anywhere
Qwen3 8B fits on a laptop GPU and still has a thinking mode. What it is good for, where a small model breaks, and why its hosted price is not what its size suggests.
ReadModelsQwen3 Coder 30B A3B Instruct: A Local Coding Agent Model
The small Qwen3 Coder: 3.3B active parameters, 256K context, built for agentic coding tools. How its tool-call format works, FIM support, and when to pick it over Instruct.
ReadModelsQwen3-Coder 480B A35B: The Open Coder That Started the Line
Qwen3-Coder-480B-A35B shipped in July 2025 as an Apache 2.0 agentic coder. What it is, how it was trained, how to run it, and whether it still beats its successor.
ReadModelsQwen3-Coder-Flash: The Cheap Qwen Coder and Its Price Tiers
Qwen3-Coder-Flash is the fast, low-cost tier of Alibaba's hosted Qwen3-Coder. What is known about it, how its tiered pricing really behaves, and the jobs it suits.
ReadShowing 253–264 of 446 articles