Models / Phi 4

Microsoft

Phi 4

19 published results from 4 sources. Each card shows where the number comes from and what it does not measure. Results are never combined into one score.

Provider
Microsoft
Sources
4
Our benchmarks
0
Price
Not yet published

Reported by others

Reported by Arena (formerly LMArena)
1,256
Business, management and finance · rank 282 of 402
Unit
Arena rating, higher is better
Range
1,244 to 1,268
Sample
2438 votes
Configuration
Phi 4
Measured
25 Sep 2026
Not shown
Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.
Reported by Arena (formerly LMArena)
1,210
Creative writing · rank 301 of 407
Unit
Arena rating, higher is better
Range
1,200 to 1,220
Sample
4062 votes
Configuration
Phi 4
Measured
25 Sep 2026
Not shown
Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.
Reported by Arena (formerly LMArena)
1,268
Expert prompts · rank 252 of 359
Unit
Arena rating, higher is better
Range
1,251 to 1,286
Sample
1124 votes
Configuration
Phi 4
Measured
25 Sep 2026
Not shown
Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.
Reported by Arena (formerly LMArena)
1,245
Instruction following · rank 295 of 409
Unit
Arena rating, higher is better
Range
1,239 to 1,252
Sample
9162 votes
Configuration
Phi 4
Measured
25 Sep 2026
Not shown
Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.
Reported by Arena (formerly LMArena)
1,256
Overall · rank 307 of 409
Unit
Arena rating, higher is better
Range
1,252 to 1,261
Sample
24126 votes
Configuration
Phi 4
Measured
25 Sep 2026
Not shown
Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.
Reported by Arena (formerly LMArena)
1,229
Writing, literature and language · rank 305 of 408
Unit
Arena rating, higher is better
Range
1,222 to 1,237
Sample
6470 votes
Configuration
Phi 4
Measured
25 Sep 2026
Not shown
Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.
Reported by Berkeley Function Calling Leaderboard (BFCL) V4
24.7%
Memory · rank 31 of 109
Unit
% correct, higher is better
Configuration
phi-4
Measured
16 Dec 2025
Not shown
Not long-term personal memory in a product: sessions are BFCL's scripted ones.
Reported by Berkeley Function Calling Leaderboard (BFCL) V4
3.9%
Multi-turn tasks · rank 84 of 109
Unit
% correct, higher is better
Configuration
phi-4
Measured
16 Dec 2025
Not shown
Not open-ended agent work: the tools and tasks are BFCL's simulated APIs.
Reported by Berkeley Function Calling Leaderboard (BFCL) V4
28.8%
Overall accuracy · rank 70 of 109
Unit
% correct, higher is better
Configuration
phi-4
Measured
16 Dec 2025
Not shown
Not a neutral average: the weighting is BFCL's. Not reliability on your own tools: BFCL's functions and queries are a fixed test set, and a correct call is judged by its form, not by what it achieved.
Reported by Berkeley Function Calling Leaderboard (BFCL) V4
50.0%
Relevance detection · rank 98 of 109
Unit
% correct, higher is better
Configuration
phi-4
Measured
16 Dec 2025
Not shown
Not whether the call itself was right; only that one was attempted.
Reported by Berkeley Function Calling Leaderboard (BFCL) V4
69.6%
Single-turn calls (curated) · rank 88 of 109
Unit
% correct, higher is better
Configuration
phi-4
Measured
16 Dec 2025
Not shown
Not reliability on your own tools: BFCL's functions and queries are a fixed test set, and a correct call is judged by its form, not by what it achieved.
Reported by Berkeley Function Calling Leaderboard (BFCL) V4
60.7%
Single-turn calls (user-contributed) · rank 85 of 109
Unit
% correct, higher is better
Configuration
phi-4
Measured
16 Dec 2025
Not shown
Not reliability on your own tools: BFCL's functions and queries are a fixed test set, and a correct call is judged by its form, not by what it achieved.
Reported by Berkeley Function Calling Leaderboard (BFCL) V4
4.5%
Web search · rank 65 of 109
Unit
% correct, higher is better
Configuration
phi-4
Measured
16 Dec 2025
Not shown
Not general research quality: questions have short, checkable answers.
Reported by UGI Leaderboard
19.0%
Requested-length error · rank 163 of 370
Unit
% off the requested word count, lower is better
Configuration
Phi 4
Measured
16 Nov 2025
Not shown
Not other format limits such as character counts or bullet counts.
Reported by UGI Leaderboard
0.34
Style adherence · rank 209 of 370
Unit
score from 0 to 1, higher is better
Configuration
Phi 4
Measured
16 Nov 2025
Not shown
Not brand voice on your own examples: UGI's prompts are private and lean towards creative writing.
Reported by UGI Leaderboard
25.7
Writing score · rank 300 of 370
Unit
score out of 100, higher is better
Configuration
Phi 4
Measured
16 Nov 2025
Not shown
Not business copy quality: UGI's prompts are private and lean towards creative writing, and models that often refuse get no score.
Reported by Vectara Hallucination Leaderboard
80.7%
Answer rate · rank 104 of 108
Unit
% of documents, higher is better
Configuration
Phi 4
Measured
22 Sep 2026
Not shown
Not a quality score: a low rate usually means content filters were triggered, and hallucination rates are measured on answered documents only.
Reported by Vectara Hallucination Leaderboard
3.7%
Hallucination rate · rank 4 of 108
Unit
% of summaries, lower is better
Configuration
Phi 4
Measured
22 Sep 2026
Not shown
Not errors in open questions or other tasks: only summarisation, judged by Vectara's own model (HHEM-2.3), not by people, on news-style documents rather than your data.

Compare Phi 4 with