06-28-2026, 07:15 PM
An AI OpenWeight copywriting model that writes like a senior direct-response copywriter, not a chatbot.
I was tired of AI writing bad copy. Did a lot of hard work and open sourced a fine tuned gemma 4 31B. Try it out if you want to its a bit hard to setup you need to be technical but 100% open weight.
https://huggingface.co/akwin123/copywriter-gemma4-31b
GGUF: https://huggingface.co/models?other=base...gemma4-31b (some one made GGUF for llama.cpp or LMSTUDIO tech)
EQ-Bench 3 as Emotional Intelligence Benchmark methodology was used to judge the AI before and after fine tuning. Free Open Source Link.
Main metric — preference Elo
Recommended decoding
Thinking: OFF
How to run this LLM?
Fireworks.ai, together.ai, vast.ai, runpod etc or other providers for serverless options are best hope. They will charge some cents per hour. or your own hardware.
Recommended GPU: NVIDIA A6000 or NVIDIA A100 or higher.
I am just a copywriter trying to make something better and openweight aka free kinda if we exclude price to run LLM. This was my first attempt. Let me know how the AI if you can run it. I was bored of other AI so i ended up finetuning with private knowledge dataset of my own not public ones.
If i ask what is copywriting to Claude.ai whose's score is higher than this LLM offcourse, I like answer of this LLM more. Needs more testing from you guys offcourse.
if you have questions ? drop them.
I was tired of AI writing bad copy. Did a lot of hard work and open sourced a fine tuned gemma 4 31B. Try it out if you want to its a bit hard to setup you need to be technical but 100% open weight.
https://huggingface.co/akwin123/copywriter-gemma4-31b
GGUF: https://huggingface.co/models?other=base...gemma4-31b (some one made GGUF for llama.cpp or LMSTUDIO tech)
- ? What it is: A Gemma-4-31B model specialized for copywriting and marketing
- ? Headline result: Elo 1657 vs 1367 for base — wins 24 of 30 head-to-head briefs (80%)
- ⚖️ Judge: DeepSeek V4 Flash, blind pairwise (both orderings)
- ✍️ Best at: Ad copy, email, landing pages, hooks, product copy, scripts
- ? Context: 256K tokens · bfloat16 · drops into vLLM / Transformers
EQ-Bench 3 as Emotional Intelligence Benchmark methodology was used to judge the AI before and after fine tuning. Free Open Source Link.
Main metric — preference Elo
- Copywriter-Gemma4-31B: 1657 ? (Wins 24 / 30 vs base - 80%)
- Gemma-4-31B-it (base): 1367
Recommended decoding
- temperature: 0.7 (lively but on-brief)
- min_p: 0.1 (trims low-probability noise)
- max_new_tokens: 256–1024 (by format)
Thinking: OFF
How to run this LLM?
Fireworks.ai, together.ai, vast.ai, runpod etc or other providers for serverless options are best hope. They will charge some cents per hour. or your own hardware.
Recommended GPU: NVIDIA A6000 or NVIDIA A100 or higher.
I am just a copywriter trying to make something better and openweight aka free kinda if we exclude price to run LLM. This was my first attempt. Let me know how the AI if you can run it. I was bored of other AI so i ended up finetuning with private knowledge dataset of my own not public ones.
If i ask what is copywriting to Claude.ai whose's score is higher than this LLM offcourse, I like answer of this LLM more. Needs more testing from you guys offcourse.
if you have questions ? drop them.