Phi-4-mini

By price, Phi-4-mini ranks 1st cheapest of 68 on input and 1st on output. Its quality score of 68 ranks 65th of 68, and on value (quality score ÷ output price) it ranks 1st of 68. Its blended rate sits 99% below the catalog average.

In real billing terms: a support chatbot handling 50,000 turns a month (2,000 input / 500 output tokens each) costs about $3.00 per month on Phi-4-mini, while a RAG workload of 20,000 queries (8,000 input / 800 output tokens each) runs about $3.84. Because output is priced 2.0× input here, long-generation tasks scale cost fastest — cap output length before you optimise anything else.

A 128K context window holds roughly 171 pages of A4 text (at ~750 tokens per page), ranking 53rd of 68 here. That covers typical RAG and multi-turn chat, though book-length inputs still need retrieval and chunking. Output runs at about 250 tok/s (2nd fastest of 68), fast enough for real-time interaction and high-throughput batches.

We track 4 Microsoft models, and Phi-4-mini's blended rate sits 77% below that provider's average. Among the models tracked here it is already one of the cheapest options at this quality level. Going the other way, qwen3.5-flash scores 82 on quality for about 120% more.

The closest model in Microsoft's line-up is Phi-4 ($0.07 input / $0.14 output, quality 76). Against it, Phi-4-mini scores 8 points lower on quality, runs 70 tok/s faster (about 39%), has a 112K larger context window, costs 250% less blended. When two models sit this close on price, the deciding factor is usually the quality ceiling — run both against your real prompts before picking.