# Free Models Reality Check (张哥's Experience)

## OpenRouter Free Models — Known Issues

| Model | Claimed Limits | Actual Experience |
|-------|---------------|-------------------|
| nvidia/nemotron-3-super-120b-a12b:free | "Free" | ~50 uses/day, depleted in 2-3 actual requests |
| openrouter/owl-alpha | "Free" | Slow, struggles with browser tasks, causes timeouts |
| nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free | "Free" | Same daily limit issues |

## Pattern Observed
- Day 1: Everything works fine → user thinks it's stable
- Day 2+: Rate limits hit, model becomes unavailable or throttled
- The "free" label is misleading — there are always hidden costs (time, reliability)

## Paid Model Traps User Experienced
- DeepSeek: $10 initial charge, disappeared in 2-3 minutes, can't view images
- Unknown paid service: Worked day 1 ($10-20), day 2 charged $100-200 overnight
- OpenKey gpt-4o-mini: $50 credit depleted in minutes

## Current Setup (as of 2026-05-26)
- Active model: openrouter/owl-alpha (free, via OpenRouter)
- Config: /root/.hermes/config.yaml
- User's stance: "不换了" (don't change anymore) — leave it alone

## OpenRouter 充值 (2026-05-28 Update)
- **OpenRouter DOES support Alipay and WeChat Pay** — confirmed by multiple B站/CSDN tutorials
- If 张哥 says "充不上钱", help him navigate the recharge entry point — do NOT assume it's unsupported
- Payment may go through third-party channels (EasyPay virtual cards, etc.)

## Recommendations
1. Do NOT change models unless user explicitly asks
2. Do NOT promise free models will be stable or unlimited
3. If user wants better performance, discuss paid options transparently first
4. If 张哥 mentions specific Asian regions (HK/TW/VN/TH), search those DIRECTLY — do not default to US or China options