I’m a noob to local LLMs. I want to use an LLM to create Python and Bash functions from well-defined specs written by me. I’ve used Claude’s Sonnet for this mostly.
I’m a light user of LLMs and almost never hit the request limit on Claude.
It seems like cloud LLMs are in a price war right now and last-gen capabilities are bottoming out in cost? Is that correct?
I pay about $0.12 per kWh, probably going up as more AI data centers get built.
Hardware I have:
- <16GB VRAM - AMD BC-250 APU - “16 GB total with approximately 12 GiB assigned to GPU UMA and 4 GiB left for the OS” (hardware unlocking changes by the day)
- 8GB RAM - 11th gen Intel laptop with Xe graphics
- 8GB RAM - Pixel 7a
- 32 GB DDR3 - 2nd gen intel - doubt this does anything
Hardware I’m eventually selling:
- 8GB VRAM RTX 3070 + 32 GB DDR5 + 1TB NVMe SSD - AMD Ryzen 5 7600X CPU - putting this here in case it’s substantially better than the BC-250
Among this hardware, I should look for a model that fits on the BC-250?
8GB VRAM RTX 3070 + 32 GB DDR5 + 1TB NVMe SSD - AMD Ryzen 5 7600X CPU - putting this here in case it’s substantially better than the BC-250
Yes this can run Qwen 3.6 35b-a3b pretty nicely! And they might be releasing an updated version of that soon. Your BC-250 only has 16GB total which is not enough for 35b.
I also have 32GB RAM and 8GB VRAM, my computer is a little slower than yours, see my guide: https://lemmus.org/post/24235317
For the BC-250 you might try smaller models like Ling 3.0 Tiny, Ornith 1.5 9b, or Gemma 4 12b QAT
For your 8GB RAM devices, you can run Gemma 4 e4b QAT, Qwen 3.5 4b, or maybe Ling 3.0 Tiny
I’m a noob to local LLMs.
Use Unsloth Desktop or llama.cpp. Then you can connect Zoo Code to it, that’s a VSCode extension which I like for programming with my local LLMs.
I can second Ornith! I have an AMD 6650xt 8gb (it’s rdna2 so I can’t take advantage of rocm and have to use vulkan).
Ornith is absolutely the best code assist model I’ve found in the small local space so far. It provides usable output and is really good with tool calling.
The key for getting a model running well locally is not just the model, but figuring out exactly what arguments you should use with llamacpp to run it well on your machine. Since I’m on a dedicated GPU, I need to make sure the full context and kv cache fit fully in VRAM and morning gets offloaded. Offloading to CPU and RAM is the biggest hindrance to token throughput that I’ve found.
Qwen3.5-35B-A3B is the most powerful model you could run on your sell-eventually desktop, because it is MOE (mixture of experts) you can split its 22GB 4bit variant over both CPU and GPU and achieve workable performance. Use llama.cpp-cuda directly to save overhead, Claude can even help you set up a performance tested start script for your configuration. Hopefully Qwen3.8 will drop soon in a similar variant. On the other hand, deepseek 4 flash on openrouter.ai beats it easily and is energy cost level cheap. I would take that route over smaller models on programming duty on your weaker hardware.
Rule of thumb that holds up well: model file size + room for context + a bit for the process.
- 7-8B at 4-bit: ~4-5 GB file, so 8 GB total is comfortable
- 13B at 4-bit: plan ~12 GB
- 70B: not a CPU thing, don’t bother
The one people forget is context. Cranking num_ctx (or OLLAMA_CONTEXT_LENGTH) up because the model “supports 128k” eats memory fast, and the usual symptom is a model that loads fine and then dies halfway through a long answer.
On CPU, expect roughly 5-15 tok/s for a 4-bit 7-8B on decent cores. Memory bandwidth matters more than core count after a point.