Model comparison
Mixtral 8x22B vs phi-3-medium 14B
phi-3-medium 14B is the stronger model overall, scoring 29.7 to 27.1 on the Noometry Index.
Last verified . 6 shared benchmarks.
Mistral AI
MicrosoftSummary
- They share 6 benchmarks with published results for both. Mixtral 8x22B scores higher in 1 category and phi-3-medium 14B in 2 categories; 3 gaps are clear of the uncertainty.
- The widest gap is in coding, where phi-3-medium 14B leads 36.8 to 24.2.
- The biggest single-benchmark swing is MATH Level 5: 24.2% for Mixtral 8x22B and 17.6% for phi-3-medium 14B.
Side by side
| Mixtral 8x22B | phi-3-medium 14B | |
|---|---|---|
| Provider | Mistral AI | Microsoft |
| Noometry Index | 27.1 | 29.7 |
| Released | 2024-04-17 | 2024-04-23 |
| Weights | Open | Open |
| Context window | 64K | — |
| Max output | 64K | — |
| Input $ / M tokens | $2 | — |
| Output $ / M tokens | $6 | — |
| Results tracked | 34 | 13 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding phi-3-medium 14B leads
Mixtral 8x22B: 24.2 (#329), phi-3-medium 14B: 36.8 (#201)
| Benchmark | Mixtral 8x22B | phi-3-medium 14B |
|---|---|---|
| BigCodeBench Instruct | 40.6% | 37.6% |
| BigCodeBench Complete | 50.2% | 48.7% |
| WeirdML | 3.2% | — |
| LMArena Coding | 1166 | — |
| HumanEval+ | 72% | — |
| MBPP+ | 64.3% | — |
Agentic & Tool Use Not comparable
Mixtral 8x22B: 23.1 (#127), phi-3-medium 14B: —
| Benchmark | Mixtral 8x22B | phi-3-medium 14B |
|---|---|---|
| Cybench | 7.5% | — |
Reasoning Not comparable
Mixtral 8x22B: 19.9 (#248), phi-3-medium 14B: —
| Benchmark | Mixtral 8x22B | phi-3-medium 14B |
|---|---|---|
| Epoch Capabilities Index | 122.03 | 121.23 |
| LMArena Hard Prompts | 1150 | — |
| DTBench | 55.1% | — |
| Adversarial NLI | — | 55.8% |
| BIG-Bench Hard | — | 81.4% |
| ForecastBench | 56.3 | — |
| HellaSwag | — | 82.4% |
| WinoGrande | — | 81.5% |
Math phi-3-medium 14B leads
Mixtral 8x22B: 22.9 (#275), phi-3-medium 14B: 27.3 (#250)
| Benchmark | Mixtral 8x22B | phi-3-medium 14B |
|---|---|---|
| MATH Level 5 | 24.2% | 17.6% |
| Omni-MATH | 16.3% | — |
| LMArena Math | 1184 | — |
Knowledge Mixtral 8x22B leads
Mixtral 8x22B: 15.1 (#293), phi-3-medium 14B: 9.1 (#306)
| Benchmark | Mixtral 8x22B | phi-3-medium 14B |
|---|---|---|
| GPQA Diamond | 34.1% | 27.6% |
| MMLU | 77.8% | 78% |
| MMLU-Pro | 46% | — |
| GPQA (HELM) | 33.4% | — |
| LMArena Expert | 1113 | — |
| ARC (AI2) Challenge | — | 91.6% |
| OpenBookQA | — | 87.4% |
| TriviaQA | — | 73.9% |
Multilingual Not comparable
Mixtral 8x22B: 32.8 (#255), phi-3-medium 14B: —
| Benchmark | Mixtral 8x22B | phi-3-medium 14B |
|---|---|---|
| LMArena Non-English | 1128 | — |
| LMArena Chinese | 1116 | — |
| LMArena French | 1166 | — |
| LMArena German | 1141 | — |
| LMArena Japanese | 1037 | — |
| LMArena Korean | 1057 | — |
| LMArena Russian | 1158 | — |
| LMArena Spanish | 1151 | — |
Instruction Following Not comparable
Mixtral 8x22B: 57.7 (#266), phi-3-medium 14B: —
| Benchmark | Mixtral 8x22B | phi-3-medium 14B |
|---|---|---|
| IFEval | 72.4% | — |
| LMArena Instruction Following | 1147 | — |
Long Context Not comparable
Mixtral 8x22B: 34.7 (#247), phi-3-medium 14B: —
| Benchmark | Mixtral 8x22B | phi-3-medium 14B |
|---|---|---|
| LMArena Longer Query | 1144 | — |
Writing & Preference Not comparable
Mixtral 8x22B: 36.9 (#262), phi-3-medium 14B: —
| Benchmark | Mixtral 8x22B | phi-3-medium 14B |
|---|---|---|
| LMArena Text | 1162 | — |
| LMArena Creative Writing | 1141 | — |
| WildBench | 71.1% | — |
| LMArena Multi-Turn | 1130 | — |
Frequently asked questions
Is Mixtral 8x22B better than phi-3-medium 14B?
phi-3-medium 14B is the stronger model overall, scoring 29.7 to 27.1 on the Noometry Index.
Is Mixtral 8x22B or phi-3-medium 14B better for coding?
phi-3-medium 14B scores higher on coding benchmarks: 36.8 versus 24.2 in the Noometry coding category.
How many benchmarks do Mixtral 8x22B and phi-3-medium 14B share?
6 benchmarks have published results for both models. Mixtral 8x22B has 34 scored results on Noometry and phi-3-medium 14B has 13.