With 15 RPM, you can handle up to ~21,600 requests per day. That's enough for a real product with moderate traffic — not just a prototype.

OpenAI: $5 Credit (Runs Out Fast)

OpenAI gives new accounts a $5 credit. Here's how long it lasts at different usage levels:

With GPT-4o ($2.50/$10 per 1M tokens), the $5 credit lasts even shorter. OpenAI's free tier is really just for evaluation — you'll need to pay within a week of any real usage.

Anthropic: $5 Credit (Similar Story)

Anthropic also offers a $5 starting credit. Claude Haiku 4.5 at $1.00/$5.00 per 1M tokens stretches further than GPT-4o but still runs out quickly:

The credit works across all Claude models, so you can test Haiku, Sonnet, and even Opus — but at Opus pricing ($15/$75), the $5 evaporates in minutes.

Mistral & Cohere: $5 Credits

Both Mistral and Cohere follow the same $5 credit model. The main reasons to use their free tiers:

• Mistral: European data sovereignty, competitive pricing, strong multilingual support
• Cohere: Best-in-class for RAG and semantic search, purpose-built enterprise features

Neither offers a meaningful long-term free tier — these credits are for evaluation only.

Together.ai: $5 for Open-Source Models

Together.ai gives $5 in free credits to access open-source models like Llama 3.1, Mixtral, and others. The pricing is already low (Llama 3.1 8B at $0.18/$0.18 per 1M tokens), so $5 goes further:

Together.ai is the best option if you specifically want to use open-source models without managing your own GPU infrastructure.

What Can You Actually Build for Free?

Truly Free (Google Gemini)

• Production chatbot — 100-500 requests/day, fully functional
• Content generation tool — blog drafts, social posts, product descriptions
• Code review bot — analyze PRs, suggest improvements
• Document Q&A system — internal knowledge base search
• Classification pipeline — categorize support tickets, emails, content

Free for Prototyping (OpenAI/Anthropic/Mistral)

• MVP validation — build and test a concept before committing to paid APIs
• Hackathon projects — $5 is plenty for a weekend hackathon
• Learning and experimentation — explore different models and prompting techniques
• Personal tools — low-usage utilities for yourself

Free Tier Limitations to Watch

Free tiers come with trade-offs beyond just usage limits:

• Rate limits: Google's 15 RPM means you can't handle traffic spikes
• No SLA: Free tiers don't come with uptime guarantees
• Data training: Some providers may use free-tier data for model training (check ToS)
• Feature restrictions: Some advanced features (function calling, system prompts) may be limited
• No priority: Paid users get priority during high-demand periods

When to Upgrade from Free

You should move to a paid plan when:

1. You exceed rate limits — Google's 15 RPM becomes a bottleneck
2. You need reliability — production apps need SLAs and consistent performance
3. You need advanced features — fine-tuning, higher context windows, batch processing
4. Your $5 credit runs out — obvious but worth stating
5. You're making money — if your app generates revenue, invest in reliable API access

— See if you're overpaying for AI APIs