Local AI news
Short updates about local model releases, practical GPU fit changes, Ollama notes, and data improvements for Can My GPU Run This LLM.
The public dataset now covers 24 local model entries and 9 hardware presets. This update adds Qwen3.6 27B, a 32GB Blackwell workstation preset for RTX 5090, and new long-tail guides for RTX 5090 local LLMs and Qwen3.6 27B GPU requirements.
The model radar review found that Ollama now describes the 8B DeepSeek R1 route as DeepSeek-R1-0528-Qwen3-8B. The calculator keeps the same practical Ollama command, ollama run deepseek-r1:8b, but the model name and agent-facing guide now reflect the updated reasoning-model label.
Gemini CLI and Devstral Small 2 were also reviewed in this pass. Gemini CLI stays on the application watchlist, while Devstral Small 2 was already covered in the 23-model dataset.
The curated model dataset now covers 23 local AI models. New entries include Qwen3.5 9B, Qwen3.5 27B, Gemma 4 E4B, and Gemma 4 31B with Ollama commands, practical Q4 VRAM estimates, capability notes, and generated model pages.
This update came from an internal model-radar review. GLM-5.1 and Nemotron 3 Super were reviewed but not added to this local GPU-fit pass because they are better treated as cloud or very-large-workstation candidates for now.
The calculator now shows a RunPod cloud GPU fallback only when the selected local hardware is clearly too small for the selected model. The compatibility result and alternative-model ranking remain data-based.
The site now includes model pages, GPU/VRAM pages, an Ollama command reference, and an expanded sitemap. Google Search Console detected 29 pages after the sitemap submission.