RunAIHow.comLearn How To Run LocallyUpdated: OCTOBER 2026
Text Gen — Model Guides
Guides to the most popular text-generating models: what they are, what they are good at, the VRAM they need, and what their licenses allow.
Kimi K3
Open weightsMoonshot AI
Moonshot's flagship trillion-parameter MoE with million-token context — frontier knowledge at agentic speed.
5 min read · Open guide →
Gemma 4 31B
Open weightsGoogle
Google's most capable single-GPU Gemma — frontier-adjacent quality that fits comfortably in 24 GB.
5 min read · Open guide →
GLM-5.3
Open weightsZ.ai
Z.ai's open flagship: 744B total parameters with 40B active — MIT-licensed frontier intelligence.
5 min read · Open guide →
Qwen3.8 27B
Open weightsAlibaba
Apache-2.0 dense model with hybrid attention — one of the best quality-per-VRAM buys on a 16–24 GB GPU.
5 min read · Open guide →
DeepSeek V4.1 Flash
Open weightsDeepSeek
DeepSeek's cost-efficiency tier: a 552B sparse MoE that beats its own Pro tier on speed and task completion.
5 min read · Open guide →
Llama 3.3 70B
Open weightsMeta
Meta's dense 70B workhorse — the classic 'two-GPU' model that made serious local AI mainstream.
5 min read · Open guide →
Also see the blog for deeper guides on hardware, quantization, and costs.