diff --git a/wizard-work/quill/free-tier-ai-stack-audit.md b/wizard-work/quill/free-tier-ai-stack-audit.md new file mode 100644 index 0000000..f6cecf9 --- /dev/null +++ b/wizard-work/quill/free-tier-ai-stack-audit.md @@ -0,0 +1,38 @@ +# Free-Tier AI Stack Audit — Draft +**Date:** 2025-10-28 +**Scope:** Providers offering zero-cost API access for LLMs (as of 2025 search results). +**Evidence:** Aggregated from publicapis.io, madappgang.com, itsfree.ai, aicalcus.com, dreamprompting.com, github.com/alihesari/awesome-free-ai-apis, github.com/Unsocial-sannyasi563/awesome-free-models. + +## Verified Free-Tier Providers (API) + +| Provider | Free Quota | Models | Notes | +|----------|-----------|--------|-------| +| Google AI Studio | 1,500 req/day, 1M tokens/min | Gemini 2.0/2.5 Flash, Flash-Lite | Requires API key | +| Groq | 30 req/min, ~14,400 req/day | LLaMA 3.3 70B, Mixtral, GPT-OSS | Ultra-fast inference | +| OpenRouter | Several models at $0 | 100+ free models (e.g. gpt-oss-20b, gemma-4-31b-it) | Unified API, key required | +| Hugging Face | Limited Inference API | 300+ open-source | Good for experimentation | +| Together AI | $1 free credit on signup | Llama, Mistral, DeepSeek | Credits expire | +| Cohere | 1,000 req/month | Command R+, Command A | Good for RAG | +| Mistral AI (La Plateforme) | 1 req/s, 500k tokens/min | Mistral Large, Nemo, Codestral | Phone verification required | +| DeepSeek | 5M free tokens | DeepSeek V3, R1 | Signup required | +| Cloudflare Workers AI | 10,000 neurons/day | Llama 3.3 70B, Gemma, Nemotron | Edge deployment | +| Cerebras | 5 RPM, 1M tokens/day, 30 days | GPT-OSS 120B, Qwen3 235B | No credit card | +| Ollama Cloud | Monthly starter credits | 19 models including Kimi K2.7, DeepSeek V4 | Local-first | +| NVIDIA NIM | ~40 RPM | 129 models (Nemotron, Llama) | Signup required | + +## Always-Free No-Account Options +- **Pollinations**: GPT-5, Claude, Gemini, DeepSeek, Llama 4 — no key needed. +- **OpenCode Free**: GPT-4o, Claude, Gemini — no auth needed. +- **Qwen Chat**: Qwen3-coder-plus/flash/next — no auth needed. +- **Kiro AI**: Claude Sonnet 4.5, Haiku 4.5, Opus 4.6 — 50 credits/month, no auth. + +## Observability Gaps +- Limits change without notice (e.g. OpenAI dropped free credits in 2025). +- Rate limits vary by model within a provider. +- "Free credit" tiers often require prepaid card to unlock after spend. + +## Recommendations +1. Start with Google AI Studio for high-volume prototyping. +2. Use Groq for speed-sensitive workloads. +3. Keep OpenRouter as a unified fallback across multiple free models. +4. Audit limits monthly — free tiers are not sticky.