phi-4

azure_ai resold GPQA 56%

Also listed as azure_ai/Phi-4, deepinfra/microsoft/phi-4, openrouter/microsoft/phi-4

Input$0.125per 1M tokens
Output$0.500per 1M tokens
Cached inputn/anot offered
A 1,000-in / 300-out request costs $0.0003 Reseller price · azure_ai · recorded 2025-02-25

Price over time

One price recorded so far — nothing to chart yet. It will appear here the day the price changes.

The same capability, cheaper

phi-4 scores 56% on GPQA Diamond. The cheapest model clearing the GPQA 50%+ bar today is qwen3-7-flash at $0.055 blended — 4.0× less than this model's $0.219. Blended = 75% input, 25% output. Capability is one benchmark, not the whole story.

56%
GPQA Diamond
14%
Mock AIME
16,384
Context window
2024-12-12
Released
Microsoft Research
Developer
1
Price changes
2025-02-25
Tracked since

Recorded price changes

One row per change, newest first. US dollars per million tokens. Each row links to the exact commit or archived page the figure was read from.

DateInputOutput Cached inBlendedChangeSource
2025-02-25$0.125$0.500$0.219litellm@57faa62

How this was collected

Prices come from replaying the version history of LiteLLM's public pricing file and from vendor pricing pages archived by the Wayback Machine, then cleaned to remove metadata edits, same-day churn, unit slips and typo corrections. The full method and every script is in the repository. List pricing, not negotiated pricing.

See how the cost of a fixed level of capability has fallen →