Microsoft · Phi-4
Phi-4
Dense 14B open small language model; no newer Phi LLM has shipped since the Phi-4 family.
- AA Intelligence
- 5.9#130 of 142
- Price
- Self-host
- Output speed
- 36tok/s
- Context window
- 16K tokens
Composite indexes
- Artificial Analysis Intelligence Index#130 of 1426
Leader: Claude Opus 5.5 at 58
Coding
- LiveCodeBench#49 of 5623.1%
Leader: Fugu at 92.9%
Agents and tool use
- τ²-Bench#83 of 880%
Leader: GLM-5.2 at 99.1%
Reasoning and knowledge
- GPQA Diamond#115 of 14257.5%
Leader: GPT-6 Astra at 96.1%
- Humanity’s Last Exam#142 of 1473.8%
Leader: Claude Opus 5.5 at 61.4%
- CritPt#113 of 1390%
Leader: GPT-5.6 Sol at 32.3%
Vision and long context
- AA Long Context Reasoning#129 of 1390%
Leader: Kimi K3 at 88.7%
Other published results
| Result | Score | Reported by | Source |
|---|---|---|---|
| aime-2025 · AIME 2025; non-reasoning | 18% | independent | Artificial Analysis: Phi-4 ↗ |
| Terminal-Bench Hard (AA) · non-reasoning | 3.8% | independent | Artificial Analysis: Phi-4 ↗ |
| IFBench · non-reasoning | 23.5% | independent | Artificial Analysis: Phi-4 ↗ |
| AA-Omniscience Index · non-reasoning | -55.7 index | independent | Artificial Analysis: Phi-4 ↗ |