It sounds like you don’t need immediate llm responses and can batch process your data nightly? Have you considered running a local llm? May not need to pay for api calls. Today’s local models are quite good. I started off with cpu and even that was fine for my pipelines.
Though haven't done any extensive testing then I personally could easily get by with current local models. The only reason I don't is that the hosted ones all have free tiers.
I started off using gpt-oss-120b on cpu. It uses about 60-65gb of memory or so but my workstation has 128gb of ram. If I had less ram, I would start off with the gpt-oss-20b model and go from there. Look for MoE models as they are more efficient to run.
My old threadripper pro was seeing about 15tps, which was quite acceptable for the background tasks I was running.
I started off using gpt-oss-120b on cpu. It uses about 60-65gb of memory or so but my workstation has 128gb of ram. If I had less ram, I would start off with the gpt-oss-20b model and go from there. Look for MoE models as they are more efficient to run.
I started off using gpt-oss-120b on cpu. It uses about 60-65gb of memory or so but my workstation has 128gb of ram. If I had less ram, I would start off with the gpt-oss-20b model and go from there. Look for MoE models as they are more efficient to run.
Do they or any other providers offer any improvements on the often-chronicled variability of quality/effort from the major two services e.g. during peak hours?
As much as I like the Claude models, they are expensive. I wouldn't use them to process large volumes of data.
Gemini 2.5 Flash-Lite is $0.10 per million tokens. Grok 4.1 Fast is really good and only $0.20. They will work just as well for most simple tasks.
Pretty straightforward. Sources dump into a queue throughout the day, regex filters the obvious junk ("lol", "thanks", bot messages never hit the LLM), then everything gets batched overnight through Anthropic's Batch API for classification. Feedback gets clustered against existing pain points or creates new ones.
Most of the cost savings came from not sending stuff to the LLM that didn't need to go there, plus the batch API is half the price of real-time calls.
It sounds like you don’t need immediate llm responses and can batch process your data nightly? Have you considered running a local llm? May not need to pay for api calls. Today’s local models are quite good. I started off with cpu and even that was fine for my pipelines.
Though haven't done any extensive testing then I personally could easily get by with current local models. The only reason I don't is that the hosted ones all have free tiers.
Agreed, I'm pretty amazed at what I'm able to do locally just with an AMD 6700XT and 32GB of RAM. It's slow, but if you've got all night...
I haven't thought about that, but really want to dig in more now. Any places you recommend starting?
I started off using gpt-oss-120b on cpu. It uses about 60-65gb of memory or so but my workstation has 128gb of ram. If I had less ram, I would start off with the gpt-oss-20b model and go from there. Look for MoE models as they are more efficient to run.
My old threadripper pro was seeing about 15tps, which was quite acceptable for the background tasks I was running.
Can you suggest any good llms for cpu?
I started off using gpt-oss-120b on cpu. It uses about 60-65gb of memory or so but my workstation has 128gb of ram. If I had less ram, I would start off with the gpt-oss-20b model and go from there. Look for MoE models as they are more efficient to run.
Following.
I started off using gpt-oss-120b on cpu. It uses about 60-65gb of memory or so but my workstation has 128gb of ram. If I had less ram, I would start off with the gpt-oss-20b model and go from there. Look for MoE models as they are more efficient to run.
This is the way. I actually mapped out the decision tree for this exact process and more here:
https://github.com/NehmeAILabs/llm-sanity-checks
That's interesting. Is there any kind of mapping to these respective models somewhere?
Yes, I included a 'Model Selection Cheat Sheet' in the README (scroll down a bit).
I map them by task type:
Tiny (<3B): Gemma 3 1B (could try 4B as well), Phi-4-mini (Good for classification). Small (8B-17B): Qwen 3 8B, Llama 4 Scout (Good for RAG/Extraction). Frontier: GPT-5, Llama 4 Maverick, GLM, Kimi
Is that what you meant?
at the sake of being obvious, do you have a tiny llm gating this decision and classifying and directing the task to its appropriate solution?
Consider using z.ai as model provider to further lower your costs.
Do they or any other providers offer any improvements on the often-chronicled variability of quality/effort from the major two services e.g. during peak hours?
This is what i was going to suggest too.
Or minimax - m2.1 release didn't make a big splash in the news, but it's really capable.
Will take a look!
As much as I like the Claude models, they are expensive. I wouldn't use them to process large volumes of data. Gemini 2.5 Flash-Lite is $0.10 per million tokens. Grok 4.1 Fast is really good and only $0.20. They will work just as well for most simple tasks.
You also can try to use cheaper models like GLM, Deepseek, Qwen,at least partially.
Have you looked into https://maartengr.github.io/BERTopic/index.html ?
consider this for addtl cost savings if local doesnt interest you - https://docs.cloud.google.com/vertex-ai/generative-ai/docs/m...
Are you also adding the proper prompt cache control attributes? I think Anthropic API still doesn't do it automatically
No I need to look into this!
Can you discuss a bit more of the architecture?
Pretty straightforward. Sources dump into a queue throughout the day, regex filters the obvious junk ("lol", "thanks", bot messages never hit the LLM), then everything gets batched overnight through Anthropic's Batch API for classification. Feedback gets clustered against existing pain points or creates new ones.
Most of the cost savings came from not sending stuff to the LLM that didn't need to go there, plus the batch API is half the price of real-time calls.