#+TITLE: Alternative LLM Providers - Subscription & Token Efficient
#+author: User
#+created: [2026-03-16 Mon 14:28]
#+DATE: 2026-03-07
#+FILETAGS: :research:llm:pricing:alternatives

* GLM-5 (Zhipu AI) - Research

** Pricing Found
- Input: $1.00 per 1M tokens
- Output: $3.20 per 1M tokens
- Context: ~744B parameters, MoE architecture
- Training: 28.5T tokens

** Comparison to Current
| Model | Input Cost | Output Cost | Context | Free Tier |
|-------|-----------|-------------|---------|-----------|
| Gemini 2.0 | $0 | $0 | 1M | ✅ Yes |
| GLM-5 | $1.00 | $3.20 | ? | ? |
| Claude | $3.00 | $15.00 | 200K | ❌ No |
| GPT-4 | varies | varies | 128K | ❌ No |

** Status: Still researching subscription/unlimited plans

* Alternative Providers to Research

** Tier 1: Subscription/Unlimited
1. *Fireworks AI* - Flat-rate inference
2. *Together AI* - Pay-per-token but high limits
3. *Replicate* - Metered but competitive
4. *Groq* - Ultra-fast, low cost

** Tier 2: Self-Hosted (One-time cost)
1. *RunPod* - GPU rental for local models
2. *Lambdalabs* - GPU cloud
3. *Local inference* - RTX 4090, etc.

** Tier 3: Open Source Providers
1. *Ollama* + RunPod/Lambda
2. *llama.cpp* quantized models
3. *vLLM* serving framework

* Research Questions

1. Does GLM-5 offer unlimited subscription tier?
2. What about Fireworks/Together flat-rate plans?
3. AWS Bedrock with flat-rate (Amazon Q)?
4. Self-hosted llama3 70B vs GLM-5 quality?

* Next Steps Needed

- Manual research required (web browsing limited)
- Check Zhipu pricing page directly
- Compare subscription tiers
- Evaluate self-hosting break-even

* Current Recommendation

*Until research complete:*
- Stay on Gemini (free tier) ✅
- Use sparingly to avoid 60/minute rate limit
- 300K tokens/day = ~9M tokens/month free

*If need more than 9M/month:* Evaluate paid tiers