This repository combats misleading "free" tiers in AI coding tools by providing a curated comparison of actual free access to pro-grade models. It details limits for tools offering GPT-5, Claude Opus/Sonnet, and Gemini Pro, benefiting engineers and researchers seeking genuine value without immediate cost.

How It Works

The project is a comparative guide, defining "pro-grade" AI models by scoring >= 60% on SWE-bench Verified. Tools are ranked by their free tier limits (requests, tokens, credits) and the specific pro-grade models accessible, offering transparency into actual utility.

Highlighted Details

• Pro-Grade Definition: Models must score >= 60% on SWE-bench Verified for practical coding capability. Qualifying models include GPT-5 (74.9%), Claude Opus 4.1 (74.5%), Claude Sonnet 4 (72.7%), Qwen3-Coder-480B (69.6%), and Gemini 2.5 Pro (63.2%).
• Generous Free Tiers: Qwen Code offers 2,000 requests/day (Qwen3-Coder-480B), Rovo Dev CLI provides 5M tokens/day (Claude Sonnet 4 beta), and Gemini CLI offers 100 requests/day (Gemini 2.5 Pro).
• Model Variety: Warp grants access to GPT-5, Claude Opus 4.1, Claude Sonnet 4, and Gemini 2.5 Pro (150 requests/month). Trae supports Claude 4 Sonnet, GPT-4.1, GPT-4o, and Gemini 2.5 Pro (10 fast + 50 slow requests/month).
• Local Model Potential: Options for local deployment (Cline, Aider, Continue.dev) offer unlimited coding assistance but require substantial RAM/VRAM.

Maintenance & Community

Community contributions via issues/PRs are crucial for updating quota/model information. A disclaimer notes no vendor affiliation and ownership of trademarks by their respective holders.

Licensing & Compatibility

No specific software license is provided for the curated list content. The project operates under a disclaimer of vendor affiliation. Compatibility for commercial use or closed-source linking is not addressed.

Limitations & Caveats

Free tier limits and model availability change frequently; accuracy is not guaranteed. Direct comparison is difficult due to varied quota systems (requests, tokens, credits). Running advanced local models requires significant hardware resources (e.g., ~150GB RAM/VRAM for Qwen3-Coder-480B GGUF).