Phi-3-mini-4k-instruct: benchmark scores, price and ranking
Phi-3-mini-4k-instruct from Microsoft ranks #536 of 743 LLMs on the LLMs Tiger aggregate benchmark score (46.2), based on 3 benchmark results captured up to 2026-10-13.
- Organisation
- Microsoft
- Type
- Open-weight / local model
- Open weights
- Yes
- Released
- 2024-04-23 (source: epoch)
- Cheapest API price
- $0.130 in / $0.520 out per million tokens (via azure_ai, observed 2026-10-13T20:47Z)
- Aggregate rank
- #536 of 743 (score 46.2)
Relative to other models, Phi-3-mini-4k-instruct is strongest on MMLU-Pro (#85 of 136) and weakest on Chess Puzzles (#119 of 122).
Across the 3 benchmarks where at least ten models are scored, Phi-3-mini-4k-instruct is in the top 10% on 0 and in the top half on 0.
Its cheapest listed input price, $0.130 per million tokens, is lower than that of 76% of the 427 models with a listed price.
Phi-3-mini-4k-instruct benchmark results
| Benchmark | Score | Rank | Source | Captured |
|---|---|---|---|---|
| MMLU-Pro | 45.7% | #85 of 136 | HuggingFace MMLU-Pro leaderboard | 2026-10-13 |
| GSM8K | 85.7% | #12 of 17 | HuggingFace GSM8K leaderboard | 2026-10-13 |
| Chess Puzzles | 0.0% | #119 of 122 | Epoch AI Benchmarking Hub | 2026-10-13 |
About Phi-3-mini-4k-instruct
Phi-3-mini-4k-instruct is an open-weight model that can run on your own hardware from Microsoft. It was released on 2024-04-23, according to Epoch AI. The cheapest API listing we track charges $0.130 per million input tokens and $0.520 per million output tokens, through azure_ai. We hold 3 benchmark results for it across 2 areas, and none is in the top tenth of its board.
Where Phi-3-mini-4k-instruct stands
On our aggregate score, Phi-3-mini-4k-instruct is #536 of 743 ranked models, in the lower half. Its score is 46.2; the scale is centred on 50, so it sits 3.8 points below the typical model. That beats 28% of the ranked field. Among the 3 ranked Microsoft models it is #1. The next one down is WizardLM-2-8x22B at 45.3.
Phi-3-mini-4k-instruct, benchmark by benchmark
Phi-3-mini-4k-instruct lands below the middle on MMLU-Pro: 45.7%, #85 of 136, with 62% of the field scoring higher. Darwin-180B-RSI leads at 88.1%; the closest result above is Yi-1.5-9B-Chat at 46.0%. It covers 57-subject knowledge + reasoning. The board median is 55.3%, so Phi-3-mini-4k-instruct is 9.6 points below it. Source: HuggingFace MMLU-Pro leaderboard, captured 2026-10-13.
Phi-3-mini-4k-instruct lands below the middle on GSM8K: 85.7%, #12 of 17, with 65% of the field scoring higher. mimo-v2.5-pro leads at 99.6%; the closest result above is internlm2_5-7b-chat at 86.0%. It covers grade-school math word problems. The board median is 86.9%, so Phi-3-mini-4k-instruct is 1.2 points below it. Source: HuggingFace GSM8K leaderboard, captured 2026-10-13.
Chess Puzzles is a weak spot. Phi-3-mini-4k-instruct scores 0.0%, #119 of 122, behind 97% of the models measured. The leader, GPT-6 Astra (max), reaches 72.0%. It covers chess tactics puzzles. The board median is 12.5%, so Phi-3-mini-4k-instruct is 12.5 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
Strengths and weaknesses
General knowledge and chat: weak, ahead of 16% of the field on average across 2 benchmarks. Its best board here is GSM8K (#12 of 17) and its weakest is Chess Puzzles (#119 of 122).
Price and value
The median input price across the 427 priced models is $0.500 per million tokens, so Phi-3-mini-4k-instruct costs less than the typical listing. Models that score within 1.5 points or better and cost less: Qwen2.5-Coder-7B-Instruct ($0.010, 49.5), DeepSeek V4.1 Flash (high) ($0.010, 48.9), Qwen2.5-Coder-3B-Instruct ($0.010, 46.5). Similar scores cost more on mimo-v2.5 ($0.140) and Magistral Small ($0.150).
Release history
The Microsoft model released before it was Phi-3-medium-128k-instruct on 2024-04-23, which scores 44.0.
Reading these numbers
The 3 scores on this page come from 3 sources: Epoch AI Benchmarking Hub (1), HuggingFace GSM8K leaderboard (1), HuggingFace MMLU-Pro leaderboard (1). No official leaderboard supplied any of them, so treat them as third-party measurements. With 3 results the aggregate rank can move a lot when one new score arrives.
Models ranked nearby
- #533 MAmmoTH2-8B-Plus (46.2)
- #534 Yi-34B (46.2)
- #535 Qwen3.6 35B-A3B (46.2)
- #537 Gemini 2.0 Flash (Feb 2025) (46.1)
- #538 Claude Sonnet 4 (unknown thinking) (46.1)
- #539 gemma2:27b-instruct-q8_0 (46.1)
More from Microsoft
- WizardLM-2-8x22B (#597, 45.3)
- Phi-3-medium-128k-instruct (#653, 44.0)
Compare it on the live leaderboard → · How the aggregate score works