Clawthor
SocialGRPO on 100M tweets

Tweet Writer

tweet-v1

Viral tweets, hooks, engagement-optimized short-form — GRPO-trained on 100M real tweets

Technical Specs

Architecture

Qwen 7B (GRPO)

Context Window

2,048 tokens

Max Output

280 tokens

Streaming

Yes

Custom System Prompts

Yes

Category

Social

Why This Model

Built on a Qwen 7B model fine-tuned with GRPO (Group Relative Policy Optimization) on 100 million real tweets. Unlike generic language models, this model learned engagement patterns directly from viral content — hook structures, ratio mechanics, and platform-native voice. The GRPO training specifically optimized for engagement signals rather than just next-token prediction. This isn’t a base model with a “write a tweet” prompt — the model internalized what makes tweets perform.

Default System Prompt

Minimal by design — the model's fine-tuning does the heavy lifting. Override with a custom system parameter if needed.

Write viral tweets. Short, punchy, engagement-optimized. No hashtags unless asked. Max 280 characters unless longer is requested.

API Usage

curl -X POST https://clawthor.xyz/api/v1/generate \
  -H "Authorization: Bearer ck_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "tweet-v1",
    "prompt": "Write a viral tweet about AI agents managing DeFi portfolios",
    "max_tokens": 280
  }'

Resources